janelia_core.dataprocessing.image_stats
Tools for efficiently calculating statistics over images.
This module is geared specifically towards working with time series of volumetric data.
William Bishop bishopw@hhmi.org
Module Contents
Functions
|
Calculates standard deviation of individual voxels through time. |
|
Creates a brain mask using the morphsnakes package. |
|
Creates a brain mask by thresholding statistics. |
|
Identifies ROIs that significantly overlap with a brain mask. |
|
Identifies ROIS that fit within a rectangular box of a set size. |
- janelia_core.dataprocessing.image_stats.std_through_time(images: list, image_slice: slice = slice(None, None, None), t_dict: dict = None, sc: pyspark.SparkContext = None, preprocess_func: Callable[[numpy.ndarray], numpy.ndarray] = None, verbose=True, correct_denom=True, h5_data_group: str = 'default') dict
Calculates standard deviation of individual voxels through time.
This function will also return the uncentered first (i.e, the mean) and second moments for each individual voxel through time as well, since these are calculated in the intermediate computations.
- Args:
images: A list of either (1) numpy arrays containing the images or (2) file paths to the image files
image_slice: The slice of each image to extract. If registration is being performed, (see t_dict), coordinates of image_slice should be for the images after registration.
t_dict: If this is not None, images will be registered before any statistics are calculated. This dictionary has two entries:
transforms: A list of registration transforms to apply to the images as they are being read in. If None, no registration will be applied.
image_shape: This is the shape of the original images being read in.
sc: If provided, spark will be used to distribute computation.
preprocess_func: A function to apply to the data of each image before calculating means. If none, a function which returns image data unchanged will be used.
verbose: True if progress updates should be printed to screen.
correct_denom: If true, standard deviation is calculated by dividing by sqrt(n - 1) where n is the number of images. If false, division by sqrt(n) is used.
h5_data_group: The hdfs data group holding image data in hdf5 files
- Returns:
A dict with the standard deviation, mean and uncentered second moments for each voxel as numpy arrays. The keys will be ‘std’, ‘mean’ and ‘sec_mom’, where ‘sec_mom’ is the uncentered second moment.
- janelia_core.dataprocessing.image_stats.create_morphsnakes_brain_mask(img: numpy.ndarray, p: numpy.ndarray, morph_params: dict = {}, verbose=True) numpy.ndarray
Creates a brain mask using the morphsnakes package.
This function operates independently on each plane.
- Args:
img: The image to find the brain mask of.
p: The percentile threshold value to use when forming the initial brain mask for each plane. Specifically, for each plane in the image, the p-th percentile is calculated. The initial mask is then formed by masking out any value less than this. p[i] contains the percentile for plane i.
morph_params: A dictionary of parameters to pass into morphological_chan_vese. See that function for more information. Reasonable initial values are:
morph_params = {‘iterations’: 50, ‘smoothing’: 4, ‘lambda1’: 1, ‘lambda2’: 1}
verbose: True if progress updates should be printed.
- Returns:
mask: A binary array the same shape as an image. Each entry indicates if the corresponding voxel in the images belongs to the brain or not.
- janelia_core.dataprocessing.image_stats.create_threshold_brain_mask(stats: dict, std_p: int = 70, mean_p: int = 70, verbose=True) numpy.ndarray
Creates a brain mask by thresholding statistics.
Given a structure of statistics produced by std_through_time, this function returns a binary array estimating which voxels belong to the brain.
The array is estimated by looking for voxels with both a mean and standard deviation which surpass a user defined threshold.
- Args:
stats: The dictionary of statistics produced by std_through_time.
std_p: The threshold percentile for standard deviation values
mean_p: The threshold percentile for mean values
verbose: True if status updates should be printed.
- Returns:
mask: A binary array the same shape as an image. Each entry indicates if the corresponding voxel in the images belongs to the brain or not.
- janelia_core.dataprocessing.image_stats.identify_rois_in_brain_mask(brain_mask: numpy.ndarray, rois: List[janelia_core.dataprocessing.roi.ROI], p_in=0.9) numpy.ndarray
Identifies ROIs that significantly overlap with a brain mask.
- Args:
brain_mask: A binary np.ndarray the same shape as the shape of images rois were extracted from, indicating which voxels belong to the brain.
rois: A list of rois objects.
p_in: The percentage of voxels of an roi that must overlap with the brain mask to be considered in the mask.
- Returns:
rois_in_brain: A np.ndarray of indices of ROIs that are in the brain mask
- janelia_core.dataprocessing.image_stats.identify_small_rois(roi_extents: numpy.ndarray, rois: List[janelia_core.dataprocessing.roi.ROI]) numpy.ndarray
Identifies ROIS that fit within a rectangular box of a set size.
- Args:
roi_extents: A 3d array of the lengths of the sides of a rectangle (in pixels) rois must fit into.
rois: A list of roi objects.
- Returns:
small_rois: A np.ndarray of indices of rois that fit within the rectangle