motionSegmentation 2.9.2

Explicit spatio-temporal regularization of motion tracking.

motionSegmentation

Explicit spatio-temporal regularization of motion tracking¹ using registered vector maps from SimpleElastix²

Installation

If your default python is python3: pip install motionSegmentation

Usage

simpleSolver(savePath,startstep=1,endstep=7,fileScale=None,getCompoundTimeList=None,compoundSchemeList=None,fftLagrangian=True,pngFileFormat=None,period=None,maskImg=True,anchor=None,peroid=None,bgrid=4.,finalShape=None,fourierTerms=4,twoD=False)

REQUIRED FILES

savePath+'/scale.txt': the length per pixel x, y, z of the image

OPTIONAL FILES

savePath+'/diastole_systole.txt': diastole and systolic time frame savePath+'/transform/manual_t{0:d}to{1:d}_0.txt'.format(anchor[n][0],anchor[n][1]): additional bspline vectors to support the motion estimation

savePath

set the folder path to be used where the generated and required files will be are.

startstep, endstep

Set the start and end step for the simpleSolver.

step 1: load image -- Image will be loaded as savePath+'/'+pngFileFormat.format(t,z)

step 2: create mask

step 3: Registrations

step 4: initialize BSF using fft

step 5: solve BSF

step 6: regrid to time

setp 7: compute compound

fileScale

set based on the scale of image dimension such that fileScale*AVERAGE_IMAGE_DIMENSION > 1

getCompoundTimeList,

set the timestep(s) for compounding in a list, if None, it will use diastolic and systolic timesteps in savePath+'/diastole_systole.txt', if file does not exist, it will be set as [0]

compoundSchemeList

set the compounding method(s) to use in a list, if None it will be set for all the supported compounding methods ['SAC','maximum','mean','wavelet']

fftLagrangian

Set to False if you do not want to use time frame 0 as reference (t -> t+1 and 0 -> t will be registered), then t -> t+1 and t -> t-1 will be registered.

pngFileFormat

set the string format to read and stacked multiple images, if None, defaults to 'time{0:03d}/slice{{0:03d}}time{0:03d}.png' for 3D and 'time{0:03d}.png' for 2D.

period

Set the period of the motion in time frame (can be a float), if None, it will default as the len(DIMENSION t)

maskImg

if True, it will auto detect and mask borders with 0 intensity

anchor

set as a List of pairs of time frame to include in the motion estimation in addition to the default: [[anchor_t_n1,anchor_t_m1],[anchor_t_n2,anchor_t_m2],...]

It will find the bspline files in savePath+'/transform/manual_t{0:d}to{1:d}_0.txt'.format(anchor[n][0],anchor[n][1]) for each pair

bgrid

set the spacing of the uniform bspline grid use to represent the image registration in terms of pixels of the largest dimension

fourierTerms

Number of Fourier terms, number of Fourier coefficients = Fourier terms*2+1

finalShape

Set the final uniform bspline grid shape (NUMBER OF CONTROL POINTS IN x,NUMBER OF CONTROL POINTS IN y,NUMBER OF CONTROL POINTS IN z,NUMBER FOURIER COEFFICIENTS PER CONTROL POINT, NUMBER OF DIMENSIONS), if None, it will default to the shape in t0 -> t1 registration.

twoD

set to True is the image set is in 2D

References

¹ Wiputra, H., Chan, W. X., Foo, Y. Y., Ho, S., & Yap, C. H. (2020). Cardiac motion estimation from medical images: a regularisation framework applied on pairwise image registration displacement fields. Scientific reports, 10(1), 1-14.

² Bradley Lowekamp, ., gabehart, ., Daniel Blezek, ., Luis Ibanez, ., Matt McCormick, ., Dave Chen, ., … Brad King, . (2015, June 26). SimpleElastix: SimpleElastix v0.9.0 (Version v0.9.0-SimpleElastix). Zenodo. http://doi.org/10.5281/zenodo.19049