def run(input_dir, phantom, suffix=".raw", output_dir="../data/simresults"):
"""
Compute the variance in three ROIs of several volumes in the given input
directory. Pass a suffix if desired.
Run this in in scripts/ if no output_dir specified.
phantom | string, prefix to output filename
"""
from constants import NOISELEVEL_FILE
OUTPUT_FNAME = NOISELEVEL_FILE.format(phantom)
raw_files = [f for f in os.listdir(input_dir) if f.endswith(suffix)]
d = OrderedDict()
for f in tqdm(sorted(raw_files)):
img = np.fromfile(
os.path.join(input_dir, f),
dtype=np.float16).reshape((250, 512, 311))
variances = compute_var_from_volume(img)
# Fetch frame and flux info from filename.
filtermode = find_filtermode(f)
frms = f[:3]
flux = f[8:13]
d[f] = (filtermode, frms, flux) + variances
if d:
outpath = os.path.join(output_dir, OUTPUT_FNAME)
# output file will be sorted because first loop runs over sorted filenames.
write_header = not os.path.exists(outpath)
with open(outpath, 'a+') as f:
if write_header:
f.write("#fmode,fmrs,flux,var1,var2,var3\n")
for vtuple in d.values():
# CSV style
f.write("{},{},{},{},{},{}\n".format(*vtuple))
EDGESHARPNESS_INPUT_DIR = VOLUME_OUTPUT_DIR
EDGESHARPNESS_OUTPUT_DIR = "../data/simresults"
EDGESHARPNESS_FILEPATH = os.path.join(
EDGESHARPNESS_OUTPUT_DIR,
EDGESHARPNESS_FILE.format(
PHANTOM_NAME+"-tv",
EDGESHARPNESS_ROI_NAME,
EDGESHARPNESS_PREFILTER_SIGMA
)
)
NOISELEVEL_OUTPUT_DIR = "../data/simresults"
NOISELEVEL_FILEPATH = os.path.join(
NOISELEVEL_OUTPUT_DIR,
NOISELEVEL_FILE.format(PHANTOM_NAME+"-tv")
)
RMSE_INPUT_DIR = PHANTOM_INPUT_DIR
RMSE_OUTPUT_DIR = "../data/simresults"
RMSE_GROUND_TRUTH_PATH = os.path.join(RMSE_INPUT_DIR, RMSE_GROUND_TRUTH_FILE)
RMSE_MASK_PATH = os.path.join(RMSE_INPUT_DIR, RMSE_MASK_FILE)
# Ground truth and mask data needs to be loaded only once.
RMSE_GROUND_TRUTH = np.fromfile(RMSE_GROUND_TRUTH_PATH, dtype=np.float32)
RMSE_MASK = np.fromfile(RMSE_MASK_PATH, dtype=np.uint8)
RMSE_FILEPATH = os.path.join(
RMSE_OUTPUT_DIR,
RMSE_FILE.format(PHANTOM_NAME+"-tv")
)
# For single test run.
def run(input_dir, phantom, save_volume,
suffix="_0filter_fdk.raw",
es_params=None,
compute_noise_level=False,
output_dir="../data/simresults",
**bf_kwargs
):
"""
Run postfilter routine on all files in input_dir that have the specified
suffix.
phantom | str, prefix for analysis files
save_volume | bool, whether to save the postfiltered volume or not
es_params | None or tuple(roi_name, sigma)
output_dir | Run this in scripts/ if not specified
bf_kwargs | sigmaISq, sigmaP, spacing, kernelScaling
"""
raw_nofilter_files = [f for f in listdir(input_dir) if f.endswith(suffix)]
COMPUTE_EDGESHARPNESS = es_params is not None
COMPUTE_ROI_VARIANCE = compute_noise_level
if COMPUTE_EDGESHARPNESS:
EDGESHARPNESS_FILEPATH = os.path.join(output_dir,
EDGESHARPNESS_FILE.format(phantom, *es_params))
ROI_VARIANCES_FILEPATH = os.path.join(output_dir, NOISELEVEL_FILE.format(phantom))
for i,file in tqdm(enumerate(raw_nofilter_files)):
# convert to float32 because pyopencl cannot handle float16.
img_in = np.fromfile(
os.path.join(input_dir, file),
dtype=np.float16).reshape((250,512,311)).astype(np.float32)
img_out = _apply_bilateral_filter(img_in, queue, **bf_kwargs)
# Analysis
if COMPUTE_ROI_VARIANCE:
variances = compute_var_from_volume(img_out)
write_header = not os.path.exists(ROI_VARIANCES_FILEPATH)
with open(ROI_VARIANCES_FILEPATH, 'a+') as rvf:
if write_header:
rvf.write("#fmode,fmrs,flux,var1,var2,var3\n")
rvf.write("{},{},{},{},{},{}\n".format(
2,
file[:3],
file[8:13],
*variances
)
)
if COMPUTE_EDGESHARPNESS:
es, esd = compute_es_from_volume(img_out, *es_params[::-1])
write_header = not os.path.exists(EDGESHARPNESS_FILEPATH)
with open(EDGESHARPNESS_FILEPATH, 'a+') as esf:
if write_header:
esf.write("#fmode,frms,flux,shrpns,uncrty\n")
esf.write("{},{},{},{},{}\n".format(
2, # post filtermode
file[:3],
file[8:13],
es,
esd
)
)
if save_volume:
# volume output directory is the input directory, i.e. on the server.
img_out.astype(np.float16).tofile(
os.path.join(
input_dir,
file.replace(suffix, "_2filter_fdk.raw")))
PHANTOM_FILE = PHANTOMS[PHANTOM_NAME]["rawfile"]
PHANTOM_FILEPATH = os.path.join(PHANTOM_INPUT_DIR, PHANTOM_FILE)
IR_SCALING = PHANTOMS[PHANTOM_NAME]["ir_scaling"]
FDK_SCALING_FILE = "fdk_postscaling.dat"
FDK_SCALING_FILEPATH = os.path.join(VOLUME_OUTPUT_DIR, FDK_SCALING_FILE)
EDGESHARPNESS_INPUT_DIR = VOLUME_OUTPUT_DIR
EDGESHARPNESS_OUTPUT_DIR = "../data/simresults"
EDGESHARPNESS_FILEPATH = os.path.join(
EDGESHARPNESS_OUTPUT_DIR,
EDGESHARPNESS_FILE.format(PHANTOM_NAME, EDGESHARPNESS_ROI_NAME, EDGESHARPNESS_PREFILTER_SIGMA),
)
NOISELEVEL_OUTPUT_DIR = "../data/simresults"
NOISELEVEL_FILEPATH = os.path.join(NOISELEVEL_OUTPUT_DIR, NOISELEVEL_FILE.format(PHANTOM_NAME))
# For single test run.
"""
PARAMETERS = {
"CBSIMULATION": {
"FRAMES": (479,),
"FLUX": (5e7,),
"NOISE": (True, )
},
"FILTERING": {
"MODE": ("no",)
},
"RECO": {
"MODE": ("FDK", ),
EDGESHARPNESS_INPUT_DIR = VOLUME_OUTPUT_DIR
EDGESHARPNESS_OUTPUT_DIR = "../data/simresults"
EDGESHARPNESS_FILEPATH = os.path.join(
EDGESHARPNESS_OUTPUT_DIR,
EDGESHARPNESS_FILE.format(
PHANTOM_NAME+"-mlem",
EDGESHARPNESS_ROI_NAME,
EDGESHARPNESS_PREFILTER_SIGMA
)
)
NOISELEVEL_OUTPUT_DIR = "../data/simresults"
NOISELEVEL_FILEPATH = os.path.join(
NOISELEVEL_OUTPUT_DIR,
NOISELEVEL_FILE.format(PHANTOM_NAME+"-mlem")
)
# For single test run.
"""
PARAMETERS = {
"CBSIMULATION": {
"FRAMES": (479,),
"FLUX": (5e7,),
"NOISE": (True, )
},
"FILTERING": {
"MODE": ("no",)
},
"RECO": {
EDGESHARPNESS_INPUT_DIR = VOLUME_OUTPUT_DIR
EDGESHARPNESS_OUTPUT_DIR = "../data/simresults"
EDGESHARPNESS_FILEPATH = os.path.join(
EDGESHARPNESS_OUTPUT_DIR,
EDGESHARPNESS_FILE.format(
PHANTOM_NAME+"-sirt",
EDGESHARPNESS_ROI_NAME,
EDGESHARPNESS_PREFILTER_SIGMA
)
)
NOISELEVEL_OUTPUT_DIR = "../data/simresults"
NOISELEVEL_FILEPATH = os.path.join(
NOISELEVEL_OUTPUT_DIR,
NOISELEVEL_FILE.format(PHANTOM_NAME+"-sirt")
)
RMSE_INPUT_DIR = PHANTOM_INPUT_DIR
RMSE_OUTPUT_DIR = "../data/simresults"
RMSE_GROUND_TRUTH_PATH = os.path.join(RMSE_INPUT_DIR, RMSE_GROUND_TRUTH_FILE)
RMSE_MASK_PATH = os.path.join(RMSE_INPUT_DIR, RMSE_MASK_FILE)
# Ground truth and mask data needs to be loaded only once.
RMSE_GROUND_TRUTH = np.fromfile(RMSE_GROUND_TRUTH_PATH, dtype=np.float32)
RMSE_MASK = np.fromfile(RMSE_MASK_PATH, dtype=np.uint8)
RMSE_FILEPATH = os.path.join(
RMSE_OUTPUT_DIR,
RMSE_FILE.format(PHANTOM_NAME+"-sirt")
)
# For single test run.
