NeuroPAL_ID Generic Methods¶

Back to overview

Legacy Method Classes¶

AutoDetect

class Methods.AutoDetect¶

Property Summary

fsize¶

options¶: Public Access

scale¶

supervoxels¶

szext¶

Method Summary

static BatchDetect(file, worm, num_neurons)¶: BATCHDETECT Batch detect neurons.

static constrain_eigenvalues(x, lb, ub)¶: Nonlinear inequality constraints (eigenvalues) for fmincon. Amin Nejat

detect(titlestr, volume, k, min_eig_thresh, scale, worm, exclusion_radius)¶

Matching pursuit algorithm for finding the best mixture of Gaussians that fits to input dataset. The algorithm runs greedy by subtracting off the brightest Gaussian at each iteration.

INPUTS¶

titlestr: a string that is shown in the progress bar volume: 4D image array (x,y,z,c), doesn’t need to be z-scored k: number of objects that the algorithm finds min_eig_thresh (in microns): if the minimum eigenvalue of the

covariance of an object is less than this the object will be removed in the procedure but is not accepted as a neuron

scale (micron per pixel): scaling between image and micron spaces exclusion_radius (in microns): no pair of objects can have distances

less that the exclusion

OUTPUTS¶

supervoxels: struct containing the locations (positions), shape: parameters (covariances), colors (colors), basline color (baseline), truncation values (trunc), and readout color from the image (color_readout)
params: struct contining the parameters chosen for running the: algorithm; parameters are half size of a neuron in microns (hnsz), number of objects (k), threshold for minimum eigenvalue (min_eig_thresh), exclusion radius (exclusion_radius)

Amin Nejat

static evaluation(image, volume, mp_params, centers, scale, precision)¶: Evaluation of the segmentation method, the output contains shape based evaluation (correlation and MSE) in the reconstruction field and location based evaluation (TP, FP, TN, FN, Accuracy, Precision, Recall, F1) in the centeres field. Amin Nejat

fit_gaussian(bpatch, colorvec, absolute_position)¶: Amin, desperately need comments to explain this method.

static gaussian_cost(x, vol, norm_factor)¶

Calculating residual between a multi-color Gaussian function and the original image.

Amin Nejat

static get_filter(sz, sigma_factor, trunc)¶

Creating a Gaussian filter 2D, 3D, or 4D for smoothing an image.

Amin Nejat

static get_gaussian(x, sz, norm)¶

Creating Gaussian function for a given set of parameters, used for visualization.

Amin Nejat

static instance()¶

AutoId

class Methods.AutoId¶

AUTO_ID Auto.

An auto_id object is has a set of possible names (different if tail/head), the method used, a likelihood matrix, and a set of ids for each of observed neurons.

Property Summary

annotation_weight¶

iter_rwc¶

iter_sinkhorn¶

max_fp¶: chosen a larger number slows down computations cubicly.

max_log_likelihood¶

min_log_likelihood¶

ncandidates¶: constants

p_fp¶: parameters for automatic detection of false positives. pf=prob of a detect being a false positive. max_fp is the maximum posible number of false positives.

theta¶

Method Summary

static BatchId(file, worm)¶: Batch ID neurons.

static MCR_solver(Y, X, sigma, lambda)¶: MCR - Multiple covariance regression solver Y - Target (n x d) X - Source (n x p) Sigma - covariances for each row of Y (d x d x n) Solving: sum_i |Y_i - X_i beta|_{sigma_i}^2

add_to_image(im, worm)¶: add_to_image function simply updates some properties of the image structure that depend on auto_id

compute_assignments()¶: compute_assignments computes some secondary properties of auto_id object

global_alignment(file, col, pos, model, annotated)¶

Global alignment based on search in the theta space

Amin & Erdem

id(file, im, worm)¶: We don’t have age-specific atlases yet. For now, use the hermaphrodite ID model to approximate anything younger adult.

static instance()¶

static isBodyID(sex, body)¶

ISBODYID Can we ID this body part? Input:

sex = the sex of the worm body = the body part we want to ID

Output:: is_ID = can we ID this body part?

static local_alignment(col, pos, model, theta, sgn, annotated)¶: local alignment between atlas and rotated points for given theta

static major_axis_align(pos, sign)¶

static pdist2_maha(X, Y, Sigma)¶: mahalanobis distance between a set of points and the components of a GMM

static rotmat(theta)¶

static scaled_rotation(X, Y)¶: Solves for Y = S*R*X + T where S is a diagonal scaling matrix R is a rotation matrix i.e. orthonormal and det(R)=1 T is a translation

static sinkhorn_logspace(logP, niter)¶

static sinkhorn_logspace_extended(logP, niter, p, n1)¶

update_confidences(im, worm)¶: update_confidences updates the confidence calculations of each manually annotated neuron, according to current alignments

update_id(im, worm)¶: update_auto_id changes the properties of auto_id given human has given the identity neuron to the neuron_i-th neuron (with the mp order). it also updates the image object accordingly.

update_log_likelihood(im, worm)¶: read the model and aligned information

static update_permutation(colors, positions, model, known)¶

static vec(x)¶: vectorize multi-dimensional array

visualize(im, worm, varargin)¶

NNDetect

class Methods.NNDetect¶

NNDETECT Detection of neural centers using neural nets

Constructor Summary

NNDetect()¶: NNDETECT Construct an instance of this class Outputs:

obj : an instance of this class

Property Summary

NN_model¶

crop_size¶

model¶

model_version¶

pst_shape¶

stride¶

Method Summary

static BatchDetect(file, worm, num_neurons)¶: BATCHDETECT Batch detect neurons.

static create_patches(data)¶

static detect(titlestr, data)¶

DETECT Use a neural network to detect neurons in the image volume. Inputs:

titlestr = a title string for the progress bar data = the image volume to search for neurons

Outputs:

supervoxels = struct containing the locations (positions), shape: parameters (covariances), colors (colors), basline color (baseline), truncation values (trunc), and readout color from the image (color_readout)
params = struct contining the parameters chosen for running the: algorithm; parameters are half size of a neuron in microns (hnsz), number of objects (k), threshold for minimum eigenvalue (min_eig_thresh), exclusion radius (exclusion_radius)

static instance()¶

static pad_to_shape(data, pst_shape)¶

static post_process(pred_p)¶

predict_nn(patches, pst_shape, titlestr)¶

static prepare_image(file)¶

static visualize(data, centers, pred_p, save, varargin)¶

Preprocess

class Methods.Preprocess¶

PREPROCESS preprocesssing tools for NeuroPAL images

Method Summary

static area_filter(volume, point, threshold, dim)¶

Heuristic filter for removing objects in a multi-color volume.

Amin Nejat

static decimate_frame(video, new_size, method)¶

decimating a multi-dimensional array DECIMATED = DECIMATE_FRAME(VIDEO, NEW_SIZE, METHOD) decimates the multi-dimensional array VIDEO to size specified by NEW_SIZE. METHOD options are:

‘linear’ ‘nearest’

Utilities¶

class Methods.Utils¶

UTILS Summary of this class goes here Detailed explanation goes here

Method Summary

static placement(sz, loc, F1)¶

Placing a subcube in a bigger cube (helper function).

Amin Nejat

static removeNearbyNeurons(neurons, min_xy, min_z)¶

REMOVENEARBYNEURONS Remove neurons that are too near each other (duplicate detections). Inputs:

neurons = the image with the list of neurons to check min_xy = the minimum distance allowed between neurons in x & y min_z = the minimum distance allowed between neurons in z

Note: we use the L2 norm for x & y distances, and L1 norm for z. We do this because x & y often have ~3x better resolution than z.

static save_for_parfor(fname, sp, mp_params)¶

static simulate_gaussian(sz, mu, sigma, props, baseline, truncation)¶

Simulate a truncated Gaussian function for fitting procedure. This function is used by matching pursuit algorithm.

Amin Nejat

static simulate_mvt(sz, mu, covariance, props, baseline, truncation)¶

Simulate a truncated Gaussian function for fitting procedure. This function is used by matching pursuit algorithm.

Amin Nejat

static sp2video(sp, sz, nsz, trunc)¶

static sub_sp(sp, subset)¶

Calculating a subset of superpixels.

Amin Nejat

static subcube(cube, loc, center)¶

Grabbing a patch of the image in a given location.

Amin Nejat

static superpose(vol, loc, F1)¶

static union_sp(sp1, sp2)¶

Calculating a subset of superpixels.

Amin Nejat

Orphaned Functions¶

Methods.logsumexp(a, dim)¶: Returns log(sum(exp(a),dim)) while avoiding numerical underflow. Default is dim = 1 (columns). logsumexp(a, 2) will sum across rows instead of columns. Unlike matlab’s “sum”, it will not switch the summing direction if you provide a row vector.

Methods.find_outlier(aligned, mu, sigma)¶

Methods.MatchHist(A)¶

Methods.run_histmatch(image_data, RGBW)¶

Methods.munkres(costMat)¶

MUNKRES Munkres (Hungarian) Algorithm for Linear Assignment Problem.

[ASSIGN,COST] = munkres(COSTMAT) returns the optimal column indices, ASSIGN assigned to each row and the minimum COST based on the assignment problem represented by the COSTMAT, where the (i,j)th element represents the cost to assign the jth job to the ith worker.

Partial assignment: This code can identify a partial assignment is a full assignment is not feasible. For a partial assignment, there are some zero elements in the returning assignment vector, which indicate un-assigned tasks. The cost returned only contains the cost of partially assigned tasks. This is vectorized implementation of the algorithm. It is the fastest among all Matlab implementations of the algorithm. Examples Example 1: a 5 x 5 example

Methods.munkres_extended(Cost, p, n1)¶: Solves an assignment problem, but accounts for possible n1 false positives, so each detect is a false positive with probability p/n1.

NeuroPAL_ID Generic Methods¶

Legacy Method Classes¶

INPUTS¶

OUTPUTS¶

Utilities¶

Orphaned Functions¶

See Also¶

Table of Contents

This Page