In brianInvo/xrayr: Static and dynamic xray analytics

Focus on chest radiography.

library( knitr )
knitr::opts_chunk$set(collapse = T, comment = "#>")

```r
library( xrayr )
library( magrittr )
library( ANTsRCore )
library( ANTsR )

Utilities

Motion correction

Motion correction.

temp1 = resampleImage( ri(1), 4 )
temp2 = resampleImage( ri(2), 4 )
tarspace = makeImage( c( dim( temp1 ), 2 ) )
pdcr = mergeListToNDImage( tarspace, list( temp1, temp2 ) )
reg = timeSeriesMotionCorrect( temp1, pdcr )

Segmentation

JLF

Denoising

Applications

Super resolution

Real data.

PROCEDURE OUTLINE

1. downsample-upsample to create simulated data
2. motion correct
3. avg/sharpen to get super-res estimate

procedure would differ slightly in a pure testing context ie it would be iterated until convergence but ultimately should probably be replaced with convnets

################
library( ANTsR )
library( xrayr )
baseDir = "/mnt/c/Users/bavants/projects/kmXrayPilotData/superRes/lowres/"
id = "0070"
fns = Sys.glob( paste0( baseDir, "*", id, "*mhd" ) )
img = antsImageRead( fns[1] )
limg = splitNDImageToList( img )
mimgLow = list()
ww = c( 5, 1:4,6:8) # arbitrarily chosen
limg = limg[ ww[ -1 ] ] # some stuff nearby
for ( k in 1:length( limg ) ) {
  # downsample by factor of X for training data
  X = 2.7
  hiSpc = antsGetSpacing( limg[[ k ]])
  loSpc = hiSpc * X
  moLo = resampleImage( limg[[ k ]], loSpc, useVoxels = F )
  mimgLow[[ k ]] = moLo
  }

# map time series
reg = timeSeriesMotionCorrect( tar, mimgLow )

ct = 1
wts = seq( from=-2, to=2, by=0.1 )
errs = rep( NA, length( wts )) 
for ( wt in wts ) {
  sup = timeSeriesSuperResolution( tar, reg, 'sharpen', param1 = wt )
  print( paste( wt,   MSE( tar, sup ) ) )
  errs[ ct ] = MSE( tar, sup )
  ct = ct + 1
  }
plot( wts, errs )

Functional studies

decomposition

References

Tanaka etc

brianInvo/xrayr documentation built on May 9, 2019, 8:13 a.m.