def get_eval_metrics(true_mask, pred_mask):
true_mask_sitk = sitk.GetImageFromArray(true_mask)
pred_mask_sitk = sitk.GetImageFromArray(pred_mask)
dsc = getDSC(true_mask_sitk, pred_mask_sitk)
h95 = getHausdorff(true_mask_sitk, pred_mask_sitk)
#vs = getVS(true_mask_sitk, pred_mask_sitk)
result = {}
result['dsc'] = dsc
result['h95'] = h95
#result['vs'] = vs
return dsc, h95 #(dsc, h95, vs)
def get_eval_metrics(true_mask, pred_mask, output_file=''):
true_mask_sitk = sitk.GetImageFromArray(true_mask)
pred_mask_sitk = sitk.GetImageFromArray(pred_mask)
dsc = getDSC(true_mask_sitk, pred_mask_sitk)
h95 = getHausdorff(true_mask_sitk, pred_mask_sitk)
vs = getVS(true_mask_sitk, pred_mask_sitk)
result = {}
result['dsc'] = dsc
result['h95'] = h95
result['vs'] = vs
if output_file != '':
with open(output_file, 'w+') as outfile:
json.dump(result, outfile)
return (dsc, h95, vs)
def test_leave_one_out(patient=0,
flair=True,
t1=True,
full=True,
first5=True,
aug=True,
verbose=False):
if patient < 20: dir = 'raw/Utrecht/'
elif patient < 40: dir = 'raw/Singapore/'
else: dir = 'raw/GE3T/'
dirs = os.listdir(dir)
dirs.sort()
dir += dirs[patient % 20]
FLAIR_image = sitk.ReadImage(dir + '/pre/FLAIR.nii.gz')
T1_image = sitk.ReadImage(dir + '/pre/T1.nii.gz')
FLAIR_array = sitk.GetArrayFromImage(FLAIR_image)
T1_array = sitk.GetArrayFromImage(T1_image)
if patient < 40: imgs_test = Utrecht_preprocessing(FLAIR_array, T1_array)
else: imgs_test = GE3T_preprocessing(FLAIR_array, T1_array)
if not flair: imgs_test = imgs_test[..., 1:2].copy()
if not t1: imgs_test = imgs_test[..., 0:1].copy()
img_shape = (rows_standard, cols_standard, flair + t1)
model = get_unet(img_shape, first5)
model_path = 'models/'
#if you want to test three models ensemble, just do like this: pred = (pred_1+pred_2+pred_3)/3
model.load_weights(model_path + str(patient) + '.h5')
pred = model.predict(imgs_test, batch_size=1, verbose=verbose)
pred[pred > 0.5] = 1.
pred[pred <= 0.5] = 0.
if patient < 40: original_pred = Utrecht_postprocessing(FLAIR_array, pred)
else: original_pred = GE3T_postprocessing(FLAIR_array, pred)
filename_resultImage = model_path + str(patient) + '.nii.gz'
sitk.WriteImage(sitk.GetImageFromArray(original_pred),
filename_resultImage)
filename_testImage = os.path.join(dir + '/wmh.nii.gz')
testImage, resultImage = getImages(filename_testImage,
filename_resultImage)
dsc = getDSC(testImage, resultImage)
avd = getAVD(testImage, resultImage)
h95 = getHausdorff(testImage, resultImage)
recall, f1 = getLesionDetection(testImage, resultImage)
return dsc, h95, avd, recall, f1
original_pred = Utrecht_postprocessing(FLAIR_array, pred)
elif np.array_equal(para_FLAIR[0], para_array[2]):
print('GE3T!')
original_pred = GE3T_postprocessing(FLAIR_array, pred)
if not os.path.exists(outputDir):
os.mkdir(outputDir)
filename_resultImage = os.path.join(outputDir, 'result.nii.gz')
sitk.WriteImage(sitk.GetImageFromArray(original_pred),
filename_resultImage)
filename_testImage = os.path.join(inputDir, dir_name, 'wmh.nii.gz')
testImage, resultImage = getImages(filename_testImage,
filename_resultImage)
dsc = getDSC(testImage, resultImage)
avd = getAVD(testImage, resultImage)
h95 = getHausdorff(testImage, resultImage)
recall, f1 = getLesionDetection(testImage, resultImage)
print('Result of patient ' + str(patient_count))
print('Dice', dsc, '(higher is better, max=1)')
print('HD', h95, 'mm', '(lower is better, min=0)')
print('AVD', avd, '%', '(lower is better, min=0)')
print('Lesion detection', recall, '(higher is better, max=1)')
print('Lesion F1', f1, '(higher is better, max=1)')
#Save result-------------------------------------------------------
result_output_dir = os.path.join(outputDir,
dir_name) #directory for images
if not os.path.exists(result_output_dir):
os.mkdir(result_output_dir)
np.save(os.path.join(result_output_dir, 'dsc.npy'), dsc)
np.save(os.path.join(result_output_dir, 'avd.npy'), avd)
np.save(os.path.join(result_output_dir, 'h95.npy'), h95)