def test_centre_of_mass(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'],
pixdim=[2])
self.assertListEqual(list(pairwise_measures.com_ref()), [0.0])
self.assertListEqual(list(pairwise_measures.com_seg()), [1.0])
self.assertEqual(pairwise_measures.com_dist(), 2.)
def test_centre_of_mass(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'],
pixdim=[2])
self.assertListEqual(list(pairwise_measures.com_ref()), [0.0])
self.assertListEqual(list(pairwise_measures.com_seg()), [1.0])
self.assertEqual(pairwise_measures.com_dist(), 2.)
def test_accuracy(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.accuracy(), 3. / 4)
def test_sensitivity(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.sensitivity(), 1.)
def test_n_negative(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.n_neg_ref(), 3.)
self.assertEqual(pairwise_measures.n_neg_seg(), 2.)
def test_faulty_inputs(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[3]['seg_img'],
ref_img=TEST_CASES[3]['ref_img'])
self.assertRaises(ValueError, pairwise_measures.tp)
def test_binary_check_for_probabilities(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[3]['seg_img'],
ref_img=TEST_CASES[3]['ref_img'])
self.assertRaises(ValueError, pairwise_measures.check_binary)
def test_informedness(self):
# true positive rate - false positive rate
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertAlmostEqual(pairwise_measures.informedness(), 2. / 3)
def test_n_negative(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.n_neg_ref(), 3.)
self.assertEqual(pairwise_measures.n_neg_seg(), 2.)
def test_true_negative(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.tn(), 2.)
def test_dice_score(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[0]['seg_img'],
ref_img=TEST_CASES[0]['ref_img'])
self.assertEqual(pairwise_measures.dice_score(), 1.0)
def test_vol_diff(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'],
pixdim=[2])
self.assertEqual(pairwise_measures.vol_diff(), 1.)
def test_markedness(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.markedness(), 1. / 2)
def test_informedness(self):
# true positive rate - false positive rate
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertAlmostEqual(pairwise_measures.informedness(), 2. / 3)
def test_intersection(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.n_intersection(), 1.)
def test_jaccard(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.jaccard(), 1. / 2)
def test_sensitivity(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.sensitivity(), 1.)
def test_markedness(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.markedness(), 1. / 2)
def test_specificity(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.specificity(), 2. / 3)
def test_vol_diff(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'],
pixdim=[2])
self.assertEqual(pairwise_measures.vol_diff(), 1.)
def test_accuracy(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.accuracy(), 3. / 4)
def test_dice_score(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[0]['seg_img'],
ref_img=TEST_CASES[0]['ref_img'])
self.assertEqual(pairwise_measures.dice_score(), 1.0)
def test_false_positive_rate(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.false_positive_rate(), 1. / 3)
def test_true_negative(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.tn(), 2.)
def test_positive_predictive_value(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.positive_predictive_values(), 1. / 2)
def test_intersection(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.n_intersection(), 1.)
def test_false_positive_rate(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.false_positive_rate(), 1. / 3)
def test_specificity(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.specificity(), 2. / 3)
def test_negative_predictive_value(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.negative_predictive_values(), 1.)
def train(dsets, model, criterion, optimizer, num_epochs, device, cp_path,
batch_size):
since = time.time()
dataloaders = {
x: DataLoader(dsets[x],
batch_size=batch_size,
shuffle=True,
num_workers=4)
for x in ['training', 'validation']
}
model = model.to(device)
for epoch in range(num_epochs):
print('Epoch {}/{}'.format(epoch + 1, num_epochs))
print('-' * 10)
# Each epoch has a training and validation phase
for phase in ['training', 'validation']:
if phase == 'training':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0.0
epoch_samples = 0
# Iterate over data
for iteration, (inputs,
labels) in enumerate(dataloaders[phase], 1):
nbatches, wsize, nchannels, x, y, z, _ = inputs.size()
inputs = inputs.view(nbatches * wsize, nchannels, x, y, z)
labels = labels.view(nbatches * wsize, nchannels, x, y, z)
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'training'):
outputs = model(inputs)
pred = (outputs > 0.5)
loss = criterion(outputs, labels)
# backward + optimize only if in training phase
if phase == 'training':
loss.backward()
optimizer.step()
# statistics
epoch_samples += inputs.size(0)
running_loss += loss.item() * inputs.size(0)
measures = PairwiseMeasures(pred.cpu().numpy(),
labels.cpu().numpy())
running_corrects += measures.dice_score() * inputs.size(0)
epoch_loss = running_loss / epoch_samples
epoch_acc = running_corrects / epoch_samples
print('{} Loss: {:.4f} Dice: {:.4f}'.format(
phase, epoch_loss, epoch_acc))
if epoch == 0:
best_loss = epoch_loss
torch.save(model.state_dict(), cp_path.format(epoch + 1))
# deep copy the model
if phase == 'validation' and epoch_loss < best_loss:
best_loss = epoch_loss
torch.save(model.state_dict(), cp_path)
print('Checkpoint {} saved!'.format(epoch + 1))
print()
time_elapsed = time.time() - since
print('Training complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
def test_jaccard(self):
pairwise_measures = PairwiseMeasures(seg_img=TEST_CASES[1]['seg_img'],
ref_img=TEST_CASES[1]['ref_img'])
self.assertEqual(pairwise_measures.jaccard(), 1. / 2)
