new_name = fname + '_pred_dice_' + str(dice_score)
save(n_results, os.path.join(save_path, new_name + '.nii'), hdr)
save(o_image, os.path.join(save_path, fname + '_img.nii'), hdr)
save(o_mask, os.path.join(save_path, fname + '_mask.nii'), hdr)
def eval_prediction(image, mask):
return model.evaluate(image, mask, verbose=1)
images_dir = './data/test/images/*'
masks_dir = './data/test/masks/*'
glob_images = sorted(glob.glob(images_dir))
glob_masks = sorted(glob.glob(masks_dir))
for file_image, file_mask in zip(glob_images, glob_masks):
original_img, img, hdr = dh.getImageData(file_image, is_eval=True)
original_msk, msk, _ = dh.getImageData(file_mask,
is_mask=True,
is_eval=True)
res = eval_prediction(img, msk)
fname = getFileName(file_image)
predict_masks(fname, img, original_img, original_msk, hdr, res[1])
print(res)
import numpy as np
import os
import skimage.io as io
from datahandler import DataHandler
from unet import getUnet
from medpy.io import save
from keras import backend as K
# get keras model
model = getUnet()
model.load_weights('unet_brain_seg_axial.h5')
# get image data and header
dh = DataHandler()
image, hdr = dh.getImageData('./data/prev/04.nii')
save_path = './data/prev/'
# predict mask
results = model.predict(image, verbose = 1)
# save mask to nifti format
# remove extra axis
n_results = np.squeeze(results)
# swap axis so they are (size, size, num slices)
n_results = np.moveaxis(n_results, 0, -1)
save(n_results, os.path.join(save_path, 'mask_04.nii'), hdr)
