def measure(self, generated, vessels, masks, num_data, iter_time, phase,
total_time):
# masking
vessels_in_mask, generated_in_mask = utils.pixel_values_in_mask(
vessels, generated, masks)
# averaging processing time
avg_pt = (total_time / num_data) * 1000 # average processing tiem
# evaluate Area Under the Curve of ROC and Precision-Recall
auc_roc = utils.AUC_ROC(vessels_in_mask, generated_in_mask)
auc_pr = utils.AUC_PR(vessels_in_mask, generated_in_mask)
# binarize to calculate Dice Coeffient
binarys_in_mask = utils.threshold_by_otsu(generated, masks)
dice_coeff = utils.dice_coefficient_in_train(vessels_in_mask,
binarys_in_mask)
acc, sensitivity, specificity = utils.misc_measures(
vessels_in_mask, binarys_in_mask)
score = auc_pr + auc_roc + dice_coeff + acc + sensitivity + specificity
# # auc_sum for saving best model in training
# auc_sum = auc_roc + auc_pr
# if self.flags.stage == 2:
# #auc_sum = auc_roc + auc_pr
# auc_sum = auc_roc + auc_pr
# else:
# auc_sum = auc_roc + auc_pr
auc_sum = dice_coeff + acc + auc_pr
# print information
ord_output = collections.OrderedDict([('auc_pr', auc_pr),
('auc_roc', auc_roc),
('dice_coeff', dice_coeff),
('acc', acc),
('sensitivity', sensitivity),
('specificity', specificity),
('score', score),
('auc_sum', auc_sum),
('best_auc_sum',
self.best_auc_sum),
('avg_pt', avg_pt)])
utils.print_metrics(iter_time, ord_output)
# write in tensorboard when in train mode only
if phase == 'train':
self.model.measure_assign(auc_pr, auc_roc, dice_coeff, acc,
sensitivity, specificity, score,
iter_time)
elif phase == 'test':
# write in npy format for evaluation
utils.save_obj(vessels_in_mask, generated_in_mask,
os.path.join(self.auc_out_dir, "auc_roc.npy"),
os.path.join(self.auc_out_dir, "auc_pr.npy"))
return auc_sum
def measure(self, generated, vessels, masks, num_data, iter_time, phase,
total_time):
vessels_in_mask, generated_in_mask = utils.pixel_values_in_mask(
vessels, generated, masks)
avg_pt = (total_time / num_data) * 1000 # average processing tiem
# evaluation
auc_roc = utils.AUC_ROC(vessels_in_mask, generated_in_mask)
auc_pr = utils.AUC_PR(vessels_in_mask, generated_in_mask)
binarys_in_mask = utils.threshold_by_otsu(generated, masks)
dice_coeff = utils.dice_coefficient_in_train(vessels_in_mask,
binarys_in_mask)
acc, sensitivity, specificity = utils.misc_measures(
vessels_in_mask, binarys_in_mask)
score = auc_pr + auc_roc + dice_coeff + acc + sensitivity + specificity
# print information
ord_output = collections.OrderedDict([('auc_pr', auc_pr),
('auc_roc', auc_roc),
('dice_coeff', dice_coeff),
('acc', acc),
('sensitivity', sensitivity),
('specificity', specificity),
('score', score),
('best_dice_coeff',
self.best_dice_coeff),
('avg_pt', avg_pt)])
utils.print_metrics(iter_time, ord_output)
# write in tensorboard
if phase == 'train':
self.model.measure_assign(auc_pr, auc_roc, dice_coeff, acc,
sensitivity, specificity, score,
iter_time)
if phase == 'test':
# write in npy format for evaluation
utils.save_obj(vessels_in_mask, generated_in_mask,
os.path.join(self.auc_out_dir, "auc_roc.npy"),
os.path.join(self.auc_out_dir, "auc_pr.npy"))
return dice_coeff
if check_validation_batch:
utils.check_input(imgs, segs, train_img_check_dir)
check_validation_batch = False
pred = network.predict(imgs, batch_size=batch_size, verbose=0)
loss = network.evaluate(imgs, segs, batch_size=batch_size, verbose=0)
losses += [loss] * imgs.shape[0]
pred_masks.append(pred)
gt_masks.append(segs)
fundus_imgs.append(imgs)
pred_masks = np.concatenate(pred_masks, axis=0)
gt_masks = np.concatenate(gt_masks, axis=0)
fundus_imgs = np.concatenate(fundus_imgs, axis=0)
# evaluate results
auroc = utils.AUC_ROC(gt_masks, pred_masks)
aupr = utils.AUC_PR(gt_masks, pred_masks)
utils.print_metrics(epoch + 1, auroc=auroc, aupr=aupr, validation_losses=np.mean(losses))
# save the weight
if aupr > best_aupr:
network.save_weights(os.path.join(model_out_dir, "network_{}.h5".format(epoch + 1)))
best_aupr = aupr
# save validation results
for index in range(pred_masks.shape[0]):
Image.fromarray((pred_masks[index, ..., 0] * 255).astype(np.uint8)).save(os.path.join(img_out_dir, str(epoch + 1) + "_{:02}_segmented.png".format(index + 1)))
Image.fromarray((gt_masks[index, ..., 0] * 255).astype(np.uint8)).save(os.path.join(img_out_dir, str(epoch + 1) + "_{:02}_gt.png".format(index + 1)))
Image.fromarray((fundus_imgs[index, ...] * 255).astype(np.uint8)).save(os.path.join(img_out_dir, str(epoch + 1) + "_{:02}_fundus_patch.png".format(index + 1)))
sys.stdout.flush()
# G
gan_x_test, gan_y_test=utils.input2gan(val_imgs, val_vessels, d_out_shape)
loss,acc=gan.evaluate(gan_x_test,gan_y_test, batch_size=batch_size, verbose=0)
utils.print_metrics(n_round+1, acc=acc, loss=loss, type='GAN')
# save the weights
g.save_weights(os.path.join(model_out_dir,"g_{}_{}_{}.h5".format(n_round,FLAGS.discriminator,FLAGS.ratio_gan2seg)))
# update step sizes, learning rates
scheduler.update_steps(n_round)
K.set_value(d.optimizer.lr, scheduler.get_lr())
K.set_value(gan.optimizer.lr, scheduler.get_lr())
# evaluate on test images
if n_round in rounds_for_evaluation:
generated=g.predict(test_imgs,batch_size=batch_size)
generated=np.squeeze(generated, axis=3)
vessels_in_mask, generated_in_mask = utils.pixel_values_in_mask(test_vessels, generated , test_masks)
auc_roc=utils.AUC_ROC(vessels_in_mask,generated_in_mask,os.path.join(auc_out_dir,"auc_roc_{}.npy".format(n_round)))
auc_pr=utils.AUC_PR(vessels_in_mask, generated_in_mask,os.path.join(auc_out_dir,"auc_pr_{}.npy".format(n_round)))
binarys_in_mask=utils.threshold_by_otsu(generated,test_masks)
dice_coeff=utils.dice_coefficient_in_train(vessels_in_mask, binarys_in_mask)
acc, sensitivity, specificity=utils.misc_measures(vessels_in_mask, binarys_in_mask)
utils.print_metrics(n_round+1, auc_pr=auc_pr, auc_roc=auc_roc, dice_coeff=dice_coeff,
acc=acc, senstivity=sensitivity, specificity=specificity, type='TESTING')
# print test images
segmented_vessel=utils.remain_in_mask(generated, test_masks)
for index in range(segmented_vessel.shape[0]):
Image.fromarray((segmented_vessel[index,:,:]*255).astype(np.uint8)).save(os.path.join(img_out_dir,str(n_round)+"_{:02}_segmented.png".format(index+1)))
total_loss, loss_all, loss_vessel = network.train_on_batch(
[imgs, vessels], [segs, segs[:, ::32, ::32, :]])
losses_all += [loss_all] * len(filenames)
losses_vessel += [loss_vessel] * len(filenames)
print "loss_all: {}, loss_vessel: {}".format(np.mean(losses_all),
np.mean(losses_vessel))
# evaluate on validation set
if check_validation_batch:
utils.check_input(val_imgs, val_masks, val_img_check_dir)
# utils.check_input(val_vessels, val_masks, val_img_check_dir)
check_test_batch = False
val_generated_masks_f_v, val_generate_masks_v = network.predict(
[val_imgs, val_vessels], batch_size=batch_size, verbose=0)
auroc = utils.AUC_ROC(val_masks, val_generated_masks_f_v)
aupr = utils.AUC_PR(val_masks, val_generated_masks_f_v)
val_generated_masks_f_v = np.round(val_generated_masks_f_v)
cm, spe, sen, dice_val, jaccard_val = utils.seg_metrics(
val_masks, val_generated_masks_f_v)
utils.print_metrics(epoch + 1,
jaccard_val=jaccard_val,
dice_val=dice_val,
sen=sen,
auroc=auroc,
aupr=aupr)
# save the weight
network.save_weights(
os.path.join(model_out_dir, "network_{}.h5".format(epoch)))
# save validation results