DeepD3 core

analysis

class deepd3.core.analysis.ROI2D_Creator(dendrite_prediction, spine_prediction, threshold)
clean(maxD, minS, dendrite_threshold=0.7)

Cleanes ROIs

Parameters

  • maxD (int) – maximum distance to dendrite in px

  • minS (int) – minimum size of ROIs in px

  • dendrite_threshold (float, optional) – _description_. Defaults to 0.7.

Returns

old ROI count, new ROI count

Return type

tuple

create(applyWatershed=False, maskSize=3, minDistance=3)

Creates 2D ROIs

Parameters

  • applyWatershed (bool, optional) – Apply Watershed algorithm. Defaults to False.

  • maskSize (int, optional) – Size of the distance transform mask. Defaults to 3.

  • minDistance (int, optional) – Minimum distance between ROIs in Watershed algorithm. Defaults to 3.

Returns

ROIs found

Return type

int

zSignal
class deepd3.core.analysis.ROI3D_Creator(dendrite_prediction, spine_prediction, mode='floodfill', areaThreshold=0.25, peakThreshold=0.8, seedDelta=0.1, distanceToSeed=10, dimensions={'xy': 0.094, 'z': 0.5})
create(minPx, maxPx, minPlanes, applyWatershed=False, dilation=0)

Create 3D ROIs

Parameters

  • minPx (int) – only retain 3D ROIs containing at least minPx pixels

  • maxPx (int) – only retain 3D ROIs containing at most maxPx pixels

  • minPlanes (int) – only retain 3D ROIs spanning at least minPlanes planes

  • applyWatershed (bool, optional) – Apply watershed algorithm to divide ROIs. Defaults to False.

  • dilation (int, optional) – Dilate dendrite probability map. Defaults to 0.

Returns

number of retained ROIs

Return type

int

log
zSignal
class deepd3.core.analysis.Stack(fn, pred_fn=None, dimensions={'xy': 0.094, 'z': 0.5})
cleanDendrite(dendrite_threshold=0.7, min_dendrite_size=100)

Cleaning dendrite

Parameters

  • dendrite_threshold (float, optional) – Dendrite probability threshold. Defaults to 0.7.

  • min_dendrite_size (int, optional) – Minimum dendrite size. Defaults to 100.

Returns

Cleaned dendrite prediction map

Return type

numpy.ndarray

cleanDendrite3D(dendrite_threshold=0.7, min_dendrite_size=100, preview=False)

Cleaning dendrites in 3D

Parameters

  • dendrite_threshold (float, optional) – Dendrite semantic segmentation threshold. Defaults to 0.7.

  • min_dendrite_size (int, optional) – Minimum dendrite size in px in 3D. Defaults to 100.

  • preview (bool, optional) – Enable preview option. Defaults to False.

Returns

Cleaned dendrite

Return type

numpy.ndarray

cleanSpines(dendrite_threshold=0.7, dendrite_dilation_iterations=12, preview=False)

Cleaning spines in 2D

Parameters

  • dendrite_threshold (float, optional) – Dendrite threshold for segmentation. Defaults to 0.7.

  • dendrite_dilation_iterations (int, optional) – Iterations to enlarge dendrite. Defaults to 12.

  • preview (bool, optional) – Enable preview option (not overwriting predictions). Defaults to False.

Returns

cleaned spines stack

Return type

numpy.ndarray

closing(iterations=1, preview=False)

Closing operation on dendrite prediction map

Parameters

  • iterations (int, optional) – Iterations of closing operation. Defaults to 1.

  • preview (bool, optional) – Enables preview mode. Defaults to False.

Returns

cleaned dendrite map

Return type

numpy.ndarray

predictFourFold(model_fn, tile_size=128, inset_size=96, pad_op=<function mean>, zmin=None, zmax=None)

Similar to predictInset (single tile prediction), but with four-way correction

Parameters

  • model_fn (str) – path to Tensorflow/Keras model

  • tile_size (int, optional) – Size of full tile. Defaults to 128.

  • inset_size (int, optional) – Size of tile inset (probability map to be kept). Defaults to 96.

  • pad_op (_type_, optional) – Padding operation. Defaults to np.mean.

  • zmin (_type_, optional) – Z-index minimum. Defaults to None.

  • zmax (_type_, optional) – Z-index maxmimum. Defaults to None.

Returns

operation was successful

Return type

bool

predictInset(model_fn, tile_size=128, inset_size=96, pad_op=<function mean>, zmin=None, zmax=None, clean_dendrite=True, dendrite_threshold=0.7)

Predict inset

Parameters

  • model_fn (str) – path to Tensorflow/Keras model

  • tile_size (int, optional) – Size of full tile. Defaults to 128.

  • inset_size (int, optional) – Size of tile inset (probability map to be kept). Defaults to 96.

  • pad_op (_type_, optional) – Padding operation. Defaults to np.mean.

  • zmin (_type_, optional) – Z-index minimum. Defaults to None.

  • zmax (_type_, optional) – Z-index maxmimum. Defaults to None.

  • clean_dendrite (bool, optional) – Cleaning dendrite. Defaults to True.

  • dendrite_threshold (float, optional) – Dendrite probability threshold. Defaults to 0.7.

Returns

operation was successful

Return type

bool

predictWholeImage(model_fn)

Predict whole image, plane by plane

Parameters

model_fn (str) – path to Tensorflow/Keras model file

Returns

operation was successful

Return type

bool

tileSignal
deepd3.core.analysis._distance_to_seed(seed, pos, delta_xy=1, delta_z=1)

Computes the euclidean distance between seed pixel and current position pos

Parameters

  • seed (tuple) – seed pixel coordinates (x,y,z)

  • pos (tuple) – current position coordinates (x,y,z)

Returns

euclidean distance between seed and current position

Return type

float

deepd3.core.analysis._get_sorted_seeds(stack, threshold=0.8)

Sort seeds according to their highest prediction value

Parameters

  • stack (numpy ndarray) – The stack with the predictions

  • threshold (float, optional) – The threshold for being a seed pixel. Defaults to 0.8.

Returns

seed coordinates sorted by prediction value

Return type

numpy.ndarray

deepd3.core.analysis._neighbours(x, y, z)

Generates 26-connected neighbours

Parameters

  • x (int) – x-value

  • y (int) – y-value

  • z (int) – z-value

Returns

neighbour indices of a given point (x,y,z)

Return type

list

deepd3.core.analysis.centroid3D(im)

Computes centroid from a 3D binary image

Parameters

im (numpy.ndarray) – binary image

Returns

centroid coordinates z,y,x

Return type

tuple

deepd3.core.analysis.centroids3D_from_labels(labels)

Computes the centroid for each label in an 3D stack containing image labels. 0 is background, 1…N are foreground labels. This function uses image moments to compute the centroid.

Parameters

labels (numpy.ndarray) – ROI labeled image (0…N)

Returns

Returns first-order moments, zero-order moments and covered planes

Return type

tuple(numpy.ndarray, numpy.ndarray, numpy.ndarray)

deepd3.core.analysis.cleanLabels(labels, rois_to_delete)

Cleans labels from label stack. Set labels in rois_to_delete to background.

Parameters

  • labels ([type]) – [description]

  • rois_to_delete ([type]) – [description]

Returns

[description]

Return type

[type]

deepd3.core.analysis.connected_components_3d(prediction, seeds, delta, threshold, distance, dimensions)

Computes connected components in 3D using various constraints. Each ROI is grown from a seed pixel. From there, in a 26-neighbour fashion more pixels are added iteratively. Each additional pixel needs to fulfill the following requirements:

  • The new pixel’s intensity needs to be in a given range relative to the seed intensity (delta)

  • The new pixel’s intensity needs to be above a given threshold

  • The new pixel’s position needs to be in the vicinity (distance) of the seed pixel

Each pixel can only be assigned to one ROI once.

Parameters

  • prediction (numpy.ndarray) – prediction from deep neural network

  • seeds (numpy.ndarray) – seed pixels

  • delta (float) – difference to seed pixel intensity

  • threshold (float) – threshold for pixel intensity

  • distance (int or float) – maximum euclidean distance in microns to seed pixel

  • dimensions (dict(float, float)) – xy and z pitch in microns

Returns

the labelled stack and the number of found ROIs

Return type

tuple(labels, N)

deepd3.core.analysis.getROIsizes(labels)

Get the ROI size for each label with one single stack pass

Parameters

labels (numpy.ndarray) – label map

Returns

the size of each ROI area

Return type

numpy.ndarray

deepd3.core.analysis.minMaxProbability(labels, prediction)

Computes the minimum and maximum probabilty of a prediction map given a label map

Parameters

  • labels (numpy.ndarray) – labels

  • prediction (numpy.ndarray) – prediction map with probabilities 0 … 1

Returns

for each label the minimum and maximum probability

Return type

numpy.ndarray

deepd3.core.analysis.reid(labels)

Relabel an existing label map to ensure continuous label ids

Parameters

labels (numpy.ndarray) – original label map

Returns

re-computed label map

Return type

numpy.ndarray

dendrite

class deepd3.core.dendrite.DendriteSWC(spacing=[1, 1, 1])
_binarize_swc_w_spheres()

Binarizes SWC file in a given 3D stack with spheres

convert(target_fn=None)

Convert swc file to tif stack

Parameters

target_fn (str, optional) – Target path. Defaults to None.

Returns

save path

Return type

return

open(swc_fn, ref_fn)

Open and read the swc and the stack file.

Parameters

  • swc_fn (str) – The file path to the swc file

  • ref_fn (str) – The file path to the stakc file

deepd3.core.dendrite.line_w_sphere(s, p0, p1, r0, r1, color=1, spacing=[1, 1, 1])

Draw a line with width in 3D space

Parameters

  • s (numpy.ndarray) – the 3D stack

  • p0 (tuple) – point 0 (x, y, z)

  • p1 (tuple) – point 1 (x, y, z)

  • r0 (float) – radius for point 0

  • r1 (float) – radius for point 1

  • color (int, optional) – Color for drawing, e.g. 255 for np.uint8 stack. Defaults to 1.

  • spacing (list, optional) – Spacing in 3D (x,y,z). Defaults to [1, 1, 1].

deepd3.core.dendrite.sphere(s, p0, d, spacing=[1, 1, 1], color=255, debug=False)

Draw a 3D sphere with given diameter d at point p0 in given color.

Parameters

  • s (numpy.ndarray) – numpy 3D stack

  • p0 (tuple) – x, y, z tuple

  • d (float) – diameter in 1 spacing unit

  • spacing (list, optional) – x, y, z spacing; x and y spacing must be equal. Defaults to [1, 1, 1].

  • color (int, optional) – Draw color, e.g. 255 for np.uint8 stack. Defaults to 255.

  • debug (bool, optional) – if True prints plane related information. Defaults to False.

deepd3.core.dendrite.xyzr(swc, i)

returns xyz coordinates and radius as tuple from swc pandas DataFrame and loc i, actually it is y, x and z

Parameters

  • swc (pandas.DataFrame) – List of traced coordinates

  • i (int) – current location

Returns

y, x, z and r coordinates as integers

Return type

tuple

spines

class deepd3.core.spines.Spines
convert()

Loads and converts spine annotation files to TIFF stacks

Returns

Path to saved TIFF stack

Return type

str

open(spines_fn: str)

Saves path to object

Parameters

spines_fn (str) – path to spines annotation file

export

class deepd3.core.export.ExportCentroids(roi_centroids)
export(fn)

Exports ROIs to file

Parameters

fn (str) – target filename and location

class deepd3.core.export.ExportFolder(rois)
export(fn, folder)

Export ROIs to folder

Parameters

  • fn (str) – file name

  • folder (str) – target folder

class deepd3.core.export.ExportImageJ(rois)
export(fn)

Export ROIs to ImageJ ROI zip file

Parameters

fn (str) – path to zip file

class deepd3.core.export.ExportPredictions(pred_spines, pred_dendrites)
export(fn, folder)

Export predictions as tif files

Parameters

  • fn (str) – file name

  • folder (str) – target folder

class deepd3.core.export.ExportROIMap(roi_map, binarize=False)
export(fn)

Export predictions as tif files

Parameters

  • fn (str) – file name

  • folder (str) – target folder

distance

deepd3.core.distance._computeDistance(p1, p2, dxy=0.1, dz=0.5)

compute euclidean distance of two points in space. Points are in (Z, Y, X) format

deepd3.core.distance._countOccurences(arr) → dict

Count occurences in array

Parameters

arr (numpy.ndarray) – Array with non-unique numbers

Returns

Dictionary with unique numbers as keys and their occurence as value

Return type

dict

deepd3.core.distance._distanceMatrix(pt1, pt2, dxy=0.1, dz=0.5) → ndarray

Compute distance matrix of points in 3D (Z, Y, X). Works only on 3D data

Parameters

  • pt1 (numpy.ndarray) – Points to be matched

  • pt2 (numpy.ndarray) – Points that can be matched

  • dxy (float, optional) – Pitch in xy. Defaults to 0.1.

  • dz (float, optional) – Pitch in z. Defaults to 0.5.

Returns

distance map from pt1 and pt2 points

Return type

numpy.ndarray

deepd3.core.distance.distanceMatrix(pt1, pt2, dxy=0.1, dz=0.5)

Compute distance matrix of points in 2D (Y, X) and 3D (Z, Y, X)

Parameters

  • pt1 (numpy.ndarray) – Points to be matched

  • pt2 (numpy.ndarray) – Points that can be matched

  • dxy (float, optional) – Pitch in xy. Defaults to 0.1.

  • dz (float, optional) – Pitch in z. Defaults to 0.5.