import os import json import csv import random import pickle import cv2 import numpy as np import torch import torch.nn as nn import torch.nn.functional as F import torch.optim as optim import torchvision.transforms as transforms from PIL import Image from torch.utils.data import Dataset, DataLoader from scipy.ndimage.measurements import label from sklearn.linear_model import LogisticRegression from sklearn.metrics import roc_auc_score, roc_curve import matplotlib.pyplot as plt import cv2 from keras.preprocessing.image import ImageDataGenerator, load_img, array_to_img, img_to_array from PIL import Image from sklearn.model_selection import train_test_split Dataset class GlaucomaDataset(Dataset): def __init__(self, root_dir, split='train', output_size=(256,256)): self.output_size = output_size self.root_dir = root_dir self.split = split self.images = [] self.segs = [] # Load data index for direct in self.root_dir: self.image_filenames = [] for path in os.listdir(os.path.join(direct, "Images_Square")): if(not path.startswith('.')): self.image_filenames.append(path) for k in range(len(self.image_filenames)): print('Loading {} image {}/{}...'.format(split, k, len(self.image_filenames)), end='\r') img_name = os.path.join(direct, "Images_Square", self.image_filenames[k]) #img = remove_nerves(np.array(Image.open(img_name).convert('RGB'))).astype(np.float32) img = np.array(Image.open(img_name).convert('RGB')) img = transforms.functional.to_tensor(img) img = transforms.functional.resize(img, output_size, interpolation=Image.BILINEAR) self.images.append(img) if split != 'test': for k in range(len(self.image_filenames)): print('Loading {} segmentation {}/{}...'.format(split, k, len(self.image_filenames)), end='\r') seg_name = os.path.join(direct, "Masks_Square", self.image_filenames[k][:-3] + "png") mask = np.array(Image.open(seg_name, mode='r')) od = (mask==1.).astype(np.float32) oc = (mask==2.).astype(np.float32) od = torch.from_numpy(od[None,:,:]) oc = torch.from_numpy(oc[None,:,:]) od = transforms.functional.resize(od, output_size, interpolation=Image.NEAREST) oc = transforms.functional.resize(oc, output_size, interpolation=Image.NEAREST) self.segs.append(torch.cat([od, oc], dim=0)) print('Succesfully loaded {} dataset.'.format(split) + ' '*50) def __len__(self): return len(self.images) def __getitem__(self, idx): img = self.images[idx] if self.split == 'test': return img else: seg = self.segs[idx] return img, seg Preprocessing def remove_nerves(image): img = array_to_img(image) img = cv2.cvtColor(np.array(img), cv2.COLOR_BGR2RGB) # convert image to grayScale grayScale = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) # kernel for morphologyEx kernel = cv2.getStructuringElement(1,(17,17)) # apply MORPH_BLACKHAT to grayScale image blackhat = cv2.morphologyEx(grayScale, cv2.MORPH_BLACKHAT, kernel) # apply thresholding to blackhat _,threshold = cv2.threshold(blackhat,10,255,cv2.THRESH_BINARY) # inpaint with original image and threshold image final_image = cv2.inpaint(img,threshold,1,cv2.INPAINT_TELEA) final_image = cv2.cvtColor(final_image, cv2.COLOR_BGR2RGB) return final_image.astype(np.float64)/255.0 Metrics EPS = 1e-7 def compute_dice_coef(input, target): ''' Compute dice score metric. ''' batch_size = input.shape[0] return sum([dice_coef_sample(input[k,:,:], target[k,:,:]) for k in range(batch_size)])/batch_size def dice_coef_sample(input, target): iflat = input.contiguous().view(-1) tflat = target.contiguous().view(-1) intersection = (iflat * tflat).sum() return (2. * intersection) / (iflat.sum() + tflat.sum()) def vertical_diameter(binary_segmentation): ''' Get the vertical diameter from a binary segmentation. The vertical diameter is defined as the "fattest" area of the binary_segmentation parameter. ''' # get the sum of the pixels in the vertical axis vertical_axis_diameter = np.sum(binary_segmentation, axis=1) # pick the maximum value diameter = np.max(vertical_axis_diameter, axis=1) # return it return diameter def vertical_cup_to_disc_ratio(od, oc): ''' Compute the vertical cup-to-disc ratio from a given labelling map. ''' # compute the cup diameter cup_diameter = vertical_diameter(oc) # compute the disc diameter disc_diameter = vertical_diameter(od) return cup_diameter / (disc_diameter + EPS) def compute_vCDR_error(pred_od, pred_oc, gt_od, gt_oc): ''' Compute vCDR prediction error, along with predicted vCDR and ground truth vCDR. ''' pred_vCDR = vertical_cup_to_disc_ratio(pred_od, pred_oc) gt_vCDR = vertical_cup_to_disc_ratio(gt_od, gt_oc) vCDR_err = np.mean(np.abs(gt_vCDR - pred_vCDR)) return vCDR_err, pred_vCDR, gt_vCDR def classif_eval(classif_preds, classif_gts): ''' Compute AUC classification score. ''' auc = roc_auc_score(classif_gts, classif_preds) return auc Post Processing def refine_seg(pred): ''' Only retain the biggest connected component of a segmentation map. ''' np_pred = pred.numpy() largest_ccs = [] for i in range(np_pred.shape[0]): labeled, ncomponents = label(np_pred[i,:,:]) bincounts = np.bincount(labeled.flat)[1:] if len(bincounts) == 0: largest_cc = labeled == 0 else: largest_cc = labeled == np.argmax(bincounts)+1 largest_cc = torch.tensor(largest_cc, dtype=torch.float32) largest_ccs.append(largest_cc) largest_ccs = torch.stack(largest_ccs) return largest_ccs Network class UNet(nn.Module): def __init__(self, n_channels=3, n_classes=2): super(UNet, self).__init__() self.n_channels = n_channels self.n_classes = n_classes self.epoch = 0 self.inc = DoubleConv(n_channels, 64) self.down1 = Down(64, 128) self.down2 = Down(128, 256) self.down3 = Down(256, 512) self.down4 = Down(512, 1024) self.down5 = Down(1024, 2048) factor = 2 self.down6 = Down(2048, 4096 // factor) self.up1 = Up(4096, 2048 // factor) self.up2 = Up(2048, 1024 // factor) self.up3 = Up(1024, 512 // factor) self.up4 = Up(512, 256 // factor) self.up5 = Up(256, 128 // factor) self.up6 = Up(128, 64) self.output_layer = OutConv(64, n_classes) def forward(self, x): x1 = self.inc(x) x2 = self.down1(x1) x3 = self.down2(x2) x4 = self.down3(x3) x5 = self.down4(x4) x6 = self.down5(x5) x7 = self.down6(x6) out = self.up1(x7, x6) out = self.up2(out, x5) out = self.up3(out, x4) out = self.up4(out, x3) out = self.up5(out, x2) out = self.up6(out, x1) out = self.output_layer(out) out = torch.sigmoid(out) return out class DoubleConv(nn.Module): """(convolution => [BN] => ReLU) * 2""" def __init__(self, in_channels, out_channels, mid_channels=None): super().__init__() if not mid_channels: mid_channels = out_channels self.double_conv = nn.Sequential( nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1), nn.BatchNorm2d(mid_channels), nn.ReLU(inplace=True), nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1), nn.BatchNorm2d(out_channels), nn.ReLU(inplace=True) ) def forward(self, x): return self.double_conv(x) class Down(nn.Module): """Downscaling with maxpool then double conv""" def __init__(self, in_channels, out_channels): super().__init__() self.maxpool_conv = nn.Sequential( nn.MaxPool2d(2), DoubleConv(in_channels, out_channels) ) def forward(self, x): return self.maxpool_conv(x) class Up(nn.Module): """Upscaling then double conv""" def __init__(self, in_channels, out_channels): super().__init__() # Use the normal convolutions to reduce the number of channels self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True) self.conv = DoubleConv(in_channels, out_channels, in_channels // 2) def forward(self, x1, x2): x1 = self.up(x1) # input is CHW diffY = x2.size()[2] - x1.size()[2] diffX = x2.size()[3] - x1.size()[3] x1 = F.pad(x1, [diffX // 2, diffX - diffX // 2, diffY // 2, diffY - diffY // 2]) x = torch.cat([x2, x1], dim=1) return self.conv(x) class OutConv(nn.Module): ''' Simple convolution. ''' def __init__(self, in_channels, out_channels): super(OutConv, self).__init__() self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1) def forward(self, x): return self.conv(x) Settings root_dirs = [ "../input/glaucoma-datasets/ORIGA","../input/glaucoma-datasets/G1020"] val_dir = [ "../input/glaucoma-datasets/REFUGE"] lr = 1e-4 batch_size = 8 num_workers = 8 total_epoch = 40 Load Data train_set = GlaucomaDataset(root_dirs, split='train') val_set = GlaucomaDataset(val_dir, split='val') train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=num_workers, pin_memory=True, ) val_loader = DataLoader(val_set, batch_size=batch_size, shuffle=False, num_workers=num_workers, pin_memory=True, ) Loading train image 13/650... /opt/conda/lib/python3.7/site-packages/ipykernel_launcher.py:23: DeprecationWarning: BILINEAR is deprecated and will be removed in Pillow 10 (2023-07-01). Use Resampling.BILINEAR instead. /opt/conda/lib/python3.7/site-packages/torchvision/transforms/functional.py:424: UserWarning: Argument interpolation should be of type InterpolationMode instead of int. Please, use InterpolationMode enum. "Argument interpolation should be of type InterpolationMode instead of int. " Loading train segmentation 13/650... /opt/conda/lib/python3.7/site-packages/ipykernel_launcher.py:34: DeprecationWarning: NEAREST is deprecated and will be removed in Pillow 10 (2023-07-01). Use Resampling.NEAREST or Dither.NONE instead. /opt/conda/lib/python3.7/site-packages/ipykernel_launcher.py:35: DeprecationWarning: NEAREST is deprecated and will be removed in Pillow 10 (2023-07-01). Use Resampling.NEAREST or Dither.NONE instead. Succesfully loaded train dataset. Succesfully loaded train dataset. Succesfully loaded val dataset. /opt/conda/lib/python3.7/site-packages/torch/utils/data/dataloader.py:490: UserWarning: This DataLoader will create 8 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary. cpuset_checked)) Init Model # Device device = torch.device("cuda:0") # Network model = UNet(n_channels=3, n_classes=2).to(device) # Loss seg_loss = torch.nn.BCELoss(reduction='mean') # Optimizer optimizer = optim.Adam(model.parameters(), lr=lr) #optim.SGD(model.parameters(), lr=0.01, momentum=0.9) Train # Define parameters nb_train_batches = len(train_loader) nb_val_batches = len(val_loader) nb_iter = 0 best_val_auc = 0. iters = list(range(1, 10)) val_losses = [] train_losses = [] train_accuracy=[] val_accuracy=[] while model.epoch < total_epoch: # Accumulators train_vCDRs, val_vCDRs = [], [] train_loss, val_loss = 0., 0. train_dsc_od, val_dsc_od = 0., 0. train_dsc_oc, val_dsc_oc = 0., 0. train_vCDR_error, val_vCDR_error = 0., 0. ############ # TRAINING # ############ model.train() train_data = iter(train_loader) for k in range(nb_train_batches): # Loads data imgs, seg_gts = train_data.next() imgs, seg_gts = imgs.to(device), seg_gts.to(device) # Forward pass logits = model(imgs) loss = seg_loss(logits, seg_gts) # Backward pass optimizer.zero_grad() loss.backward() optimizer.step() train_loss += loss.item() / nb_train_batches #for printing the loss curves train_losses.append(train_loss) with torch.no_grad(): # Compute segmentation metric pred_od = refine_seg((logits[:,0,:,:]>=0.5).type(torch.int8).cpu()).to(device) #pred_od = refine_seg((logits[:,0,:,:]>=0.5).type(torch.int8).cpu()) pred_oc = refine_seg((logits[:,1,:,:]>=0.5).type(torch.int8).cpu()).to(device) #pred_oc = refine_seg((logits[:,1,:,:]>=0.5).type(torch.int8).cpu()) gt_od = seg_gts[:,0,:,:].type(torch.int8) gt_oc = seg_gts[:,1,:,:].type(torch.int8) dsc_od = compute_dice_coef(pred_od, gt_od) dsc_oc = compute_dice_coef(pred_oc, gt_oc) train_dsc_od += dsc_od.item()/nb_train_batches train_dsc_oc += dsc_oc.item()/nb_train_batches # Compute and store vCDRs vCDR_error, pred_vCDR, gt_vCDR = compute_vCDR_error(pred_od.cpu().numpy(), pred_oc.cpu().numpy(), gt_od.cpu().numpy(), gt_oc.cpu().numpy()) train_vCDRs += pred_vCDR.tolist() train_vCDR_error += vCDR_error / nb_train_batches # Increase iterations nb_iter += 1 # Std out print('Epoch {}, iter {}/{}, loss {:.6f}'.format(model.epoch+1, k+1, nb_train_batches, loss.item()) + ' '*20, end='\r') ############## # VALIDATION # ############## model.eval() with torch.no_grad(): val_data = iter(val_loader) for k in range(nb_val_batches): # Loads data imgs, seg_gts = val_data.next() imgs, seg_gts = imgs.to(device), seg_gts.to(device) # Forward pass logits = model(imgs) val_loss += seg_loss(logits, seg_gts).item() / nb_val_batches val_losses.append(val_loss) # Std out print('Validation iter {}/{}'.format(k+1, nb_val_batches) + ' '*50, end='\r') # Compute segmentation metric pred_od = refine_seg((logits[:,0,:,:]>=0.5).type(torch.int8).cpu()).to(device) pred_oc = refine_seg((logits[:,1,:,:]>=0.5).type(torch.int8).cpu()).to(device) gt_od = seg_gts[:,0,:,:].type(torch.int8) gt_oc = seg_gts[:,1,:,:].type(torch.int8) dsc_od = compute_dice_coef(pred_od, gt_od) dsc_oc = compute_dice_coef(pred_oc, gt_oc) val_dsc_od += dsc_od.item()/nb_val_batches val_dsc_oc += dsc_oc.item()/nb_val_batches vCDR_error, pred_vCDR, gt_vCDR = compute_vCDR_error(pred_od.cpu().numpy(), pred_oc.cpu().numpy(), gt_od.cpu().numpy(), gt_oc.cpu().numpy()) val_vCDRs += pred_vCDR.tolist() val_vCDR_error += vCDR_error / nb_val_batches print('VALIDATION epoch {}'.format(model.epoch+1)+' '*50) print('LOSSES: {:.4f} (train), {:.4f} (val)'.format(train_loss, val_loss)) print('OD segmentation (Dice Score): {:.4f} (train), {:.4f} (val)'.format(train_dsc_od, val_dsc_od)) print('OC segmentation (Dice Score): {:.4f} (train), {:.4f} (val)'.format(train_dsc_oc, val_dsc_oc)) print('vCDR error: {:.4f} (train), {:.4f} (val)'.format(train_vCDR_error, val_vCDR_error)) # Save model if best validation AUC is reached if val_dsc_od + val_dsc_oc > best_val_auc: torch.save(model.state_dict(), '/kaggle/working/best_seg.pth') best_val_auc = val_dsc_od + val_dsc_oc print('Best validation AUC reached. Saved model weights.') print('_'*50) # End of epoch model.epoch += 1 VALIDATION epoch 1 LOSSES: 0.4995 (train), 0.3994 (val) OD segmentation (Dice Score): 0.3531 (train), 0.5370 (val) OC segmentation (Dice Score): 0.1653 (train), 0.1160 (val) vCDR error: 8.6617 (train), 5.7519 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 2 LOSSES: 0.3392 (train), 0.2836 (val) OD segmentation (Dice Score): 0.5450 (train), 0.6851 (val) OC segmentation (Dice Score): 0.3129 (train), 0.6201 (val) vCDR error: 3.9439 (train), 1.9759 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 3 LOSSES: 0.2276 (train), 0.2009 (val) OD segmentation (Dice Score): 0.6889 (train), 0.7708 (val) OC segmentation (Dice Score): 0.5606 (train), 0.7326 (val) vCDR error: 1.1119 (train), 0.3612 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 4 LOSSES: 0.1568 (train), 0.1576 (val) OD segmentation (Dice Score): 0.7143 (train), 0.7459 (val) OC segmentation (Dice Score): 0.5964 (train), 0.7713 (val) vCDR error: 1.1536 (train), 0.1621 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 5 LOSSES: 0.1106 (train), 0.1067 (val) OD segmentation (Dice Score): 0.7425 (train), 0.7667 (val) OC segmentation (Dice Score): 0.6068 (train), 0.7499 (val) vCDR error: 0.8591 (train), 0.2202 (val) __________________________________________________ VALIDATION epoch 6 LOSSES: 0.0792 (train), 0.0680 (val) OD segmentation (Dice Score): 0.7625 (train), 0.7374 (val) OC segmentation (Dice Score): 0.6251 (train), 0.7435 (val) vCDR error: 0.7833 (train), 0.3436 (val) __________________________________________________ VALIDATION epoch 7 LOSSES: 0.0577 (train), 0.0519 (val) OD segmentation (Dice Score): 0.7798 (train), 0.7981 (val) OC segmentation (Dice Score): 0.6356 (train), 0.8089 (val) vCDR error: 0.6444 (train), 0.1099 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 8 LOSSES: 0.0445 (train), 0.0424 (val) OD segmentation (Dice Score): 0.7932 (train), 0.6702 (val) OC segmentation (Dice Score): 0.6434 (train), 0.7996 (val) vCDR error: 0.6738 (train), 0.2564 (val) __________________________________________________ VALIDATION epoch 9 LOSSES: 0.0356 (train), 0.0324 (val) OD segmentation (Dice Score): 0.8012 (train), 0.8065 (val) OC segmentation (Dice Score): 0.6482 (train), 0.8091 (val) vCDR error: 0.7382 (train), 0.1586 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 10 LOSSES: 0.0289 (train), 0.0310 (val) OD segmentation (Dice Score): 0.8188 (train), 0.6857 (val) OC segmentation (Dice Score): 0.6598 (train), 0.7167 (val) vCDR error: 0.7524 (train), 2.2238 (val) __________________________________________________ VALIDATION epoch 11 LOSSES: 0.0241 (train), 0.0263 (val) OD segmentation (Dice Score): 0.8251 (train), 0.7683 (val) OC segmentation (Dice Score): 0.6648 (train), 0.6091 (val) vCDR error: 0.7852 (train), 0.3976 (val) __________________________________________________ VALIDATION epoch 12 LOSSES: 0.0205 (train), 0.0204 (val) OD segmentation (Dice Score): 0.8310 (train), 0.7964 (val) OC segmentation (Dice Score): 0.6720 (train), 0.8165 (val) vCDR error: 0.8110 (train), 0.1195 (val) __________________________________________________ VALIDATION epoch 13 LOSSES: 0.0174 (train), 0.0178 (val) OD segmentation (Dice Score): 0.8414 (train), 0.7825 (val) OC segmentation (Dice Score): 0.6784 (train), 0.8078 (val) vCDR error: 0.7701 (train), 0.1931 (val) __________________________________________________ VALIDATION epoch 14 LOSSES: 0.0151 (train), 0.0148 (val) OD segmentation (Dice Score): 0.8468 (train), 0.8240 (val) OC segmentation (Dice Score): 0.6851 (train), 0.8420 (val) vCDR error: 0.7495 (train), 0.0883 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 15 LOSSES: 0.0133 (train), 0.0146 (val) OD segmentation (Dice Score): 0.8527 (train), 0.7773 (val) OC segmentation (Dice Score): 0.6914 (train), 0.8128 (val) vCDR error: 0.8001 (train), 0.5041 (val) __________________________________________________ VALIDATION epoch 16 LOSSES: 0.0117 (train), 0.0138 (val) OD segmentation (Dice Score): 0.8605 (train), 0.7872 (val) OC segmentation (Dice Score): 0.7016 (train), 0.8331 (val) vCDR error: 0.7869 (train), 0.1009 (val) __________________________________________________ VALIDATION epoch 17 LOSSES: 0.0107 (train), 0.0122 (val) OD segmentation (Dice Score): 0.8630 (train), 0.7978 (val) OC segmentation (Dice Score): 0.7012 (train), 0.8533 (val) vCDR error: 0.7680 (train), 0.1034 (val) __________________________________________________ VALIDATION epoch 18 LOSSES: 0.0097 (train), 0.0116 (val) OD segmentation (Dice Score): 0.8669 (train), 0.7917 (val) OC segmentation (Dice Score): 0.7068 (train), 0.8393 (val) vCDR error: 0.8058 (train), 0.2036 (val) __________________________________________________ VALIDATION epoch 19 LOSSES: 0.0088 (train), 0.0113 (val) OD segmentation (Dice Score): 0.8725 (train), 0.7765 (val) OC segmentation (Dice Score): 0.7117 (train), 0.8286 (val) vCDR error: 0.8312 (train), 0.1506 (val) __________________________________________________ VALIDATION epoch 20 LOSSES: 0.0083 (train), 0.0097 (val) OD segmentation (Dice Score): 0.8694 (train), 0.8167 (val) OC segmentation (Dice Score): 0.7160 (train), 0.8142 (val) vCDR error: 0.8498 (train), 0.1001 (val) __________________________________________________ VALIDATION epoch 21 LOSSES: 0.0080 (train), 0.0103 (val) OD segmentation (Dice Score): 0.8677 (train), 0.7832 (val) OC segmentation (Dice Score): 0.7117 (train), 0.8348 (val) vCDR error: 0.8541 (train), 0.1551 (val) __________________________________________________ VALIDATION epoch 22 LOSSES: 0.0070 (train), 0.0085 (val) OD segmentation (Dice Score): 0.8807 (train), 0.8248 (val) OC segmentation (Dice Score): 0.7225 (train), 0.8454 (val) vCDR error: 0.8814 (train), 0.1015 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 23 LOSSES: 0.0066 (train), 0.0078 (val) OD segmentation (Dice Score): 0.8823 (train), 0.8498 (val) OC segmentation (Dice Score): 0.7273 (train), 0.8385 (val) vCDR error: 0.8874 (train), 0.1055 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 24 LOSSES: 0.0061 (train), 0.0090 (val) OD segmentation (Dice Score): 0.8862 (train), 0.7946 (val) OC segmentation (Dice Score): 0.7344 (train), 0.8406 (val) vCDR error: 0.9295 (train), 0.0886 (val) __________________________________________________ VALIDATION epoch 25 LOSSES: 0.0057 (train), 0.0114 (val) OD segmentation (Dice Score): 0.8897 (train), 0.7426 (val) OC segmentation (Dice Score): 0.7387 (train), 0.7514 (val) vCDR error: 0.9121 (train), 0.2715 (val) __________________________________________________ VALIDATION epoch 26 LOSSES: 0.0055 (train), 0.0085 (val) OD segmentation (Dice Score): 0.8911 (train), 0.8229 (val) OC segmentation (Dice Score): 0.7441 (train), 0.8232 (val) vCDR error: 0.9411 (train), 0.1522 (val) __________________________________________________ VALIDATION epoch 27 LOSSES: 0.0055 (train), 0.0097 (val) OD segmentation (Dice Score): 0.8844 (train), 0.8031 (val) OC segmentation (Dice Score): 0.7364 (train), 0.7679 (val) vCDR error: 0.9211 (train), 0.1534 (val) __________________________________________________ VALIDATION epoch 28 LOSSES: 0.0050 (train), 0.0075 (val) OD segmentation (Dice Score): 0.8933 (train), 0.8256 (val) OC segmentation (Dice Score): 0.7437 (train), 0.8468 (val) vCDR error: 0.9988 (train), 0.1010 (val) __________________________________________________ VALIDATION epoch 29 LOSSES: 0.0048 (train), 0.0086 (val) OD segmentation (Dice Score): 0.8950 (train), 0.7839 (val) OC segmentation (Dice Score): 0.7550 (train), 0.8421 (val) vCDR error: 0.9836 (train), 0.1214 (val) __________________________________________________ VALIDATION epoch 30 LOSSES: 0.0044 (train), 0.0083 (val) OD segmentation (Dice Score): 0.9009 (train), 0.7993 (val) OC segmentation (Dice Score): 0.7592 (train), 0.8405 (val) vCDR error: 0.9153 (train), 0.1565 (val) __________________________________________________ VALIDATION epoch 31 LOSSES: 0.0042 (train), 0.0074 (val) OD segmentation (Dice Score): 0.9030 (train), 0.8235 (val) OC segmentation (Dice Score): 0.7630 (train), 0.8437 (val) vCDR error: 1.0067 (train), 0.1419 (val) __________________________________________________ VALIDATION epoch 32 LOSSES: 0.0038 (train), 0.0070 (val) OD segmentation (Dice Score): 0.9108 (train), 0.8288 (val) OC segmentation (Dice Score): 0.7752 (train), 0.8519 (val) vCDR error: 1.0172 (train), 0.1210 (val) __________________________________________________ VALIDATION epoch 33 LOSSES: 0.0036 (train), 0.0083 (val) OD segmentation (Dice Score): 0.9136 (train), 0.8010 (val) OC segmentation (Dice Score): 0.7757 (train), 0.8498 (val) vCDR error: 1.0102 (train), 0.1273 (val) __________________________________________________ VALIDATION epoch 34 LOSSES: 0.0035 (train), 0.0084 (val) OD segmentation (Dice Score): 0.9144 (train), 0.8056 (val) OC segmentation (Dice Score): 0.7776 (train), 0.8487 (val) vCDR error: 0.9639 (train), 0.1004 (val) __________________________________________________ VALIDATION epoch 35 LOSSES: 0.0034 (train), 0.0083 (val) OD segmentation (Dice Score): 0.9153 (train), 0.8036 (val) OC segmentation (Dice Score): 0.7778 (train), 0.8525 (val) vCDR error: 0.9778 (train), 0.1126 (val) __________________________________________________ VALIDATION epoch 36 LOSSES: 0.0033 (train), 0.0067 (val) OD segmentation (Dice Score): 0.9164 (train), 0.8362 (val) OC segmentation (Dice Score): 0.7824 (train), 0.8564 (val) vCDR error: 0.9830 (train), 0.1085 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 37 LOSSES: 0.0031 (train), 0.0081 (val) OD segmentation (Dice Score): 0.9194 (train), 0.8076 (val) OC segmentation (Dice Score): 0.7868 (train), 0.8473 (val) vCDR error: 1.0167 (train), 0.1434 (val) __________________________________________________ VALIDATION epoch 38 LOSSES: 0.0028 (train), 0.0070 (val) OD segmentation (Dice Score): 0.9258 (train), 0.8465 (val) OC segmentation (Dice Score): 0.7911 (train), 0.8469 (val) vCDR error: 1.0796 (train), 0.0924 (val) Best validation AUC reached. Saved model weights. __________________________________________________ VALIDATION epoch 39 LOSSES: 0.0030 (train), 0.0082 (val) OD segmentation (Dice Score): 0.9216 (train), 0.8264 (val) OC segmentation (Dice Score): 0.7894 (train), 0.8499 (val) vCDR error: 1.0004 (train), 0.1101 (val) __________________________________________________ VALIDATION epoch 40 LOSSES: 0.0030 (train), 0.0093 (val) OD segmentation (Dice Score): 0.9214 (train), 0.7946 (val) OC segmentation (Dice S core): 0.7873 (train), 0.8345 (val) vCDR error: 0.9929 (train), 0.1277 (val)