def vol_proc(vol_data,
crop=None,
resize_shape=None, # None (to not resize), or vector. If vector, third entry can be None
interp_order=None,
rescale=None,
rescale_prctle=None,
resize_slices=None,
resize_slices_dim=None,
offset=None,
clip=None,
extract_nd=None, # extracts a particular section
force_binary=None, # forces anything > 0 to be 1
permute=None):
''' process a volume with a series of intensity rescale, resize and crop rescale'''
if offset is not None:
vol_data = vol_data + offset
# intensity normalize data .* rescale
if rescale is not None:
vol_data = np.multiply(vol_data, rescale)
if rescale_prctle is not None:
# print("max:", np.max(vol_data.flat))
# print("test")
rescale = np.percentile(vol_data.flat, rescale_prctle)
# print("rescaling by 1/%f" % (rescale))
vol_data = np.multiply(vol_data.astype(float), 1/rescale)
if resize_slices is not None:
resize_slices = [*resize_slices]
assert resize_shape is None, "if resize_slices is given, resize_shape has to be None"
resize_shape = resize_slices
if resize_slices_dim is None:
resize_slices_dim = np.where([f is None for f in resize_slices])[0]
assert len(resize_slices_dim) == 1, "Could not find dimension or slice resize"
resize_slices_dim = resize_slices_dim[0]
resize_shape[resize_slices_dim] = vol_data.shape[resize_slices_dim]
# resize (downsample) matrices
if resize_shape is not None and resize_shape != vol_data.shape:
resize_shape = [*resize_shape]
# allow for the last entry to be None
if resize_shape[-1] is None:
resize_ratio = np.divide(resize_shape[0], vol_data.shape[0])
resize_shape[-1] = np.round(resize_ratio * vol_data.shape[-1]).astype('int')
resize_ratio = np.divide(resize_shape, vol_data.shape)
vol_data = scipy.ndimage.interpolation.zoom(vol_data, resize_ratio, order=interp_order)
# crop data if necessary
if crop is not None:
vol_data = nd.volcrop(vol_data, crop=crop)
# needs to be last to guarantee clip limits.
# For e.g., resize might screw this up due to bicubic interpolation if it was done after.
if clip is not None:
vol_data = np.clip(vol_data, clip[0], clip[1])
if extract_nd is not None:
vol_data = vol_data[np.ix_(*extract_nd)]
if force_binary:
vol_data = (vol_data > 0).astype(float)
# return with checks. this check should be right at the end before rturn
if clip is not None:
assert np.max(vol_data) <= clip[1], "clip failed"
assert np.min(vol_data) >= clip[0], "clip failed"
return vol_data
def vol_proc(
vol_data,
crop=None,
resize_shape=None, # None (to not resize), or vector. If vector, third entry can be None
interp_order=None,
rescale=None,
rescale_prctle=None,
resize_slices=None,
resize_slices_dim=None,
offset=None,
clip=None,
extract_nd=None, # extracts a particular section
force_binary=None, # forces anything > 0 to be 1
permute=None):
''' process a volume with a series of intensity rescale, resize and crop rescale'''
if offset is not None:
vol_data = vol_data + offset
# intensity normalize data .* rescale
if rescale is not None:
vol_data = np.multiply(vol_data, rescale)
if rescale_prctle is not None:
# print("max:", np.max(vol_data.flat))
# print("test")
rescale = np.percentile(vol_data.flat, rescale_prctle)
# print("rescaling by 1/%f" % (rescale))
vol_data = np.multiply(vol_data.astype(float), 1 / rescale)
if resize_slices is not None:
resize_slices = [*resize_slices]
assert resize_shape is None, "if resize_slices is given, resize_shape has to be None"
resize_shape = resize_slices
if resize_slices_dim is None:
resize_slices_dim = np.where([f is None for f in resize_slices])[0]
assert len(resize_slices_dim
) == 1, "Could not find dimension or slice resize"
resize_slices_dim = resize_slices_dim[0]
resize_shape[resize_slices_dim] = vol_data.shape[resize_slices_dim]
# resize (downsample) matrices
if resize_shape is not None and resize_shape != vol_data.shape:
resize_shape = [*resize_shape]
# allow for the last entry to be None
if resize_shape[-1] is None:
resize_ratio = np.divide(resize_shape[0], vol_data.shape[0])
resize_shape[-1] = np.round(resize_ratio *
vol_data.shape[-1]).astype('int')
resize_ratio = np.divide(resize_shape, vol_data.shape)
vol_data = scipy.ndimage.interpolation.zoom(vol_data,
resize_ratio,
order=interp_order)
# crop data if necessary
if crop is not None:
vol_data = nd.volcrop(vol_data, crop=crop)
# needs to be last to guarantee clip limits.
# For e.g., resize might screw this up due to bicubic interpolation if it was done after.
if clip is not None:
vol_data = np.clip(vol_data, clip[0], clip[1])
if extract_nd is not None:
vol_data = vol_data[np.ix_(*extract_nd)]
if force_binary:
vol_data = (vol_data > 0).astype(float)
# return with checks. this check should be right at the end before rturn
if clip is not None:
assert np.max(vol_data) <= clip[1], "clip failed"
assert np.min(vol_data) >= clip[0], "clip failed"
return vol_data
