dataset['name'])
# evaluation
ssim_patient = mriutils.ssim(output, target)
rmse_patient = mriutils.rmse(output, target)
ssim_eval_dataset.append(ssim_patient)
rmse_eval_dataset.append(rmse_patient)
print(" Patient {:d}: {:8.4f} {:8.4f}".format(
patient, rmse_patient, ssim_patient))
output_path = '%s/%s/%d/' % (eval_output_dir,
dataset['name'], patient)
mriutils.saveAsMat(
output,
'%s/vn-%d.mat' % (output_path, epoch),
'result_vn',
mat_dict={'normalization': np.asarray(normalization)})
mriutils.saveAsMat(
target, (output_path, epoch),
'%s/reference.mat',
'reference',
mat_dict={'normalization': np.asarray(normalization)})
mriutils.saveAsMat(
input0, (output_path, epoch),
'%s/zero_filling.mat',
'result_zf',
mat_dict={'normalization': np.asarray(normalization)})
print(" Dataset {:s}: {:8.4f} {:8.4f}".format(
dataset['name'], np.mean(rmse_eval_dataset),
u_i = u_i * norm # renormalize
u_volume.append(u_i)
# postprocessing
u_volume = mriutils.postprocess(np.asarray(u_volume),
data_config['dataset']['name'])
time_reco = time.time() - eval_start_time
print('reconstruction of {} image took {:f}s'.format(
u_volume.shape, time_reco))
print('saving reconstructed image to "{}"'.format(output_name))
# save mat file
patient_id = '%s-p%d' % (data_config['dataset']['name'],
data_config['dataset']['patient'])
mriutils.saveAsMat(u_volume,
'%s-vn-%s' % (output_name, patient_id),
'result_vn',
mat_dict={'normalization': np.asarray(norm)})
# enhance volume
v_min, v_max = mriutils.getContrastStretchingLimits(
np.abs(u_volume), saturated_pixel=0.002)
volume_enhanced = mriutils.normalize(np.abs(u_volume),
v_min=v_min,
v_max=v_max)
# save pngs
for i in range(1, num_slices + 1):
mriutils.imsave(volume_enhanced[i - 1],
'%s-vn-%s-sl%d.png' % (output_name, patient_id, i))
ssim_zf = mriutils.ssim(zf_volume, target_volume)
print(
"Zero filling: RMSE={:.4f} SSIM={:.4f} VN: RMSE={:.4f} SSIM={:.4f}"
.format(rmse_zf, ssim_zf, rmse_vn, ssim_vn))
time_reco = time.time() - eval_start_time
print('reconstruction of {} image took {:f}s'.format(
u_volume.shape, time_reco))
print('saving reconstructed image to "{}"'.format(output_name))
# save mat files
patient_id = '%s-p%d' % (data_config['dataset']['name'],
data_config['dataset']['patient'])
mriutils.saveAsMat(u_volume,
'%s-vn-%s' % (output_name, patient_id),
'result_vn',
mat_dict={'normalization': np.asarray(norm)})
mriutils.saveAsMat(zf_volume,
'%s-zf-%s' % (output_name, patient_id),
'result_zf',
mat_dict={'normalization': np.asarray(norm)})
mriutils.saveAsMat(target_volume,
'%s-ref-%s' % (output_name, patient_id),
'reference',
mat_dict={'normalization': np.asarray(norm)})
# enhance image with same parameters for all images
v_min, v_max = mriutils.getContrastStretchingLimits(
np.abs(target_volume), saturated_pixel=0.002)
target_enhanced = mriutils.normalize(np.abs(target_volume),
v_min=v_min,