HDF5Array
and DelayedArray
are convenient Bioconductor packages to work with arrays “on disk”
instead of “in memory”. The cytomapper package builds upon
these tools to allow storing image data on disk. While this facilitates
the handling of hundreds to thousand of images in parallel, little
changes are experienced from the user perspective. Here, we explain
which cytomapper function are effected by storing images on
disk.
The loadImages function takes extra arguments to specify
if images should be stored on disk (on_disk) and where to
store them (h5FilesPath). When images should be stored for
longer than the current R session, the h5FilesPath needs to
be set to a permanent directory. The HDF5Array package
provides the getHDF5DumpDir function, which initialize a
temporary directory, which will be deleted once the session closes. This
is what we will use here for demonstration purposes.
library(HDF5Array)
# Define output directory
cur_dir <- getHDF5DumpDir()
path.to.images <- system.file("extdata", package = "cytomapper")
image.list <- loadImages(path.to.images, pattern = "mask.tiff",
on_disk = TRUE, h5FilesPath = cur_dir)
# Show list
image.list## CytoImageList containing 3 image(s)
## names(3): E34_mask G01_mask J02_mask
## Each image contains 1 channel## <100 x 100> DelayedMatrix object of type "double":
## [,1] [,2] [,3] ... [,99] [,100]
## [1,] 824 824 824 . 1265 1265
## [2,] 824 824 824 . 1265 0
## [3,] 824 824 824 . 0 0
## [4,] 824 824 824 . 0 1295
## [5,] 824 824 824 . 0 1295
## ... . . . . . .
## [96,] 835 0 876 . 0 0
## [97,] 835 0 876 . 0 0
## [98,] 835 0 876 . 1293 1293
## [99,] 0 0 876 . 1293 1293
## [100,] 0 0 0 . 1293 1293This function call reads in the .tiff images before writing them as
.h5 files to the indicated directory. It generates a
CytoImageList object that contains HDF5Array
or DelayedArray objects (instead of Image
objects) in each slot, which references the data in the .h5 files. The
name of the array within the .h5 file is automatically set as the
original filename and cannot be changed easily from within R. Writing
the images to disk is slow and therefore less efficient compared to
keeping images in memory. However, when working with hundreds of images
in parallel, all images remain accessible from within the R session if
they are stored on disk. In conclusion: when working with small image
sets it is recommended reading them into memory
(on_disk = FALSE, default), while large image sets should
be written to disk (on_disk = TRUE). When reading in the
same images multiple times, the .h5 files will always be replaced.
Please follow the main vignette for instructions on how to work with multi-channel images in R.
Existing CytoImageList objects, which contain individual
Image objects in memory can be converted into
CytoImageList objects storing DelayedArray or
HDF5Array objects on disk. For this the following function
calls can be used:
data("pancreasImages")
pancreasImages_onDisk <- CytoImageList(pancreasImages,
on_disk = TRUE,
h5FilesPath = cur_dir)
# Image object
pancreasImages$E34_imc## Image
## colorMode : Grayscale
## storage.mode : double
## dim : 100 100 5
## frames.total : 5
## frames.render: 5
##
## imageData(object)[1:5,1:6,1]
## [,1] [,2] [,3] [,4] [,5] [,6]
## [1,] 2.2357869 0.2537275 1.269632e+00 0.9991982 1.990020e+00 0.000000e+00
## [2,] 2.8855283 1.9900196 2.264642e+00 0.0000000 1.410924e+00 5.654589e-16
## [3,] 3.4009433 0.9950098 9.950098e-01 2.1800663 4.152935e-17 1.990020e+00
## [4,] 3.2238317 3.1750760 1.128341e+00 4.4866042 7.371460e-16 0.000000e+00
## [5,] 0.9987666 1.9900196 2.644036e-15 0.0000000 0.000000e+00 1.523360e+00## <100 x 100 x 5> HDF5Array object of type "double":
## ,,H3
## [,1] [,2] [,3] ... [,99] [,100]
## [1,] 2.2357869 0.2537275 1.2696325 . 7.265561 1.975094
## [2,] 2.8855283 1.9900196 2.2646422 . 2.985029 2.885528
## ... . . . . . .
## [99,] 2.985029 3.636569 22.976585 . 25.371881 12.045588
## [100,] 3.061645 2.985029 1.990020 . 13.353615 5.636652
##
## ...
##
## ,,CDH
## [,1] [,2] [,3] ... [,99] [,100]
## [1,] 0.940284 19.651394 43.284626 . 2.704231 0.000000
## [2,] 8.393239 29.353861 22.249359 . 7.345311 5.781126
## ... . . . . . .
## [99,] 0.9897148 7.3378549 0.0000000 . 0.000000 1.667201
## [100,] 5.8091230 2.2515676 1.9969171 . 4.344125 0.000000## An object of class "HDF5ArraySeed"
## Slot "filepath":
## [1] "/tmp/RtmpYbW74E/HDF5Array_dump/E34_imc.h5"
##
## Slot "name":
## [1] "/E34_imc"
##
## Slot "as_sparse":
## [1] FALSE
##
## Slot "type":
## [1] NA
##
## Slot "dim":
## [1] 100 100 5
##
## Slot "chunkdim":
## [1] 100 100 5
##
## Slot "first_val":
## [1] 2.235787## [1] "1.2 Mb"## [1] "11.8 Kb"Images can also be moved back to in memory representation:
pancreasImages_inMemory <- CytoImageList(pancreasImages_onDisk,
on_disk = FALSE)
# Compare the image data to the original representation
identical(as.list(pancreasImages_inMemory), as.list(pancreasImages))## [1] TRUEWhile most functions of the cytomapper package natively
support images stored on disk, there are three exceptions: the
normalize, setChannels and
mergeChannels functions.
The normalize function will store the normalized images
as a second dataset in the same .h5 file as the original data.
## [1] "11.8 Kb"## [1] 148.147## An object of class "HDF5ArraySeed"
## Slot "filepath":
## [1] "/tmp/RtmpYbW74E/HDF5Array_dump/E34_imc.h5"
##
## Slot "name":
## [1] "/E34_imc_norm"
##
## Slot "as_sparse":
## [1] FALSE
##
## Slot "type":
## [1] NA
##
## Slot "dim":
## [1] 100 100 5
##
## Slot "chunkdim":
## [1] 100 100 5
##
## Slot "first_val":
## [1] 0.01235403## [1] "11.8 Kb"## [1] 414.936As we can see, the size in memory does not increase when normalizing
images. However, the size on disk increases since a second, normalized
dataset is stored in the .h5 file. The original dataset can be
overwritten by setting overwrite = TRUE to save space on
disk. This will however break the links to the original data in all R
objects. It is therefore recommended leaving the default
overwrite = FALSE. Furthermore, the normalization of images
stored on disk is slower compared to normalizing images in memory since
normalized images need to be written to disk.
The setChannels function replaces the same channels in
all images by a user defined channel. There is no problem with this when
keeping images in memory. However, the DelayedArray
framework stores the replacement value in subassignemt operations in
memory. This means that when using the setChannels
function, the size of the object increases in memory usage:
cur_Images1 <- pancreasImages_onDisk
cur_Images2 <- getChannels(pancreasImages_onDisk, 2)
channelNames(cur_Images2) <- "CD99_2"
setChannels(cur_Images1, 1) <- cur_Images2
format(object.size(cur_Images1), units = "auto")## [1] "27.2 Kb"The mergeChannels function merges multiple user-defined
channels. As this operation creates a completely new image object, one
needs to store the merged channels in a different location:
channels1 <- getChannels(pancreasImages_onDisk, 1:2)
channels2 <- getChannels(pancreasImages_onDisk, 3:4)
dir.create(file.path(cur_dir, "test"))
cur_path_2 <- file.path(cur_dir, "test")
channels3 <- mergeChannels(channels1, channels2,
h5FilesPath = cur_path_2)
seed(channels3$E34_imc)## An object of class "HDF5ArraySeed"
## Slot "filepath":
## [1] "/tmp/RtmpYbW74E/HDF5Array_dump/test/E34_imc.h5"
##
## Slot "name":
## [1] "/E34_imc"
##
## Slot "as_sparse":
## [1] FALSE
##
## Slot "type":
## [1] NA
##
## Slot "dim":
## [1] 100 100 4
##
## Slot "chunkdim":
## [1] 100 100 4
##
## Slot "first_val":
## [1] 2.235787## R version 4.4.1 (2024-06-14)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: Etc/UTC
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] HDF5Array_1.33.8 rhdf5_2.49.0
## [3] DelayedArray_0.31.14 SparseArray_1.5.45
## [5] S4Arrays_1.5.11 abind_1.4-8
## [7] Matrix_1.7-1 ggplot2_3.5.1
## [9] cowplot_1.1.3 cytomapper_1.19.0
## [11] SingleCellExperiment_1.27.2 SummarizedExperiment_1.35.5
## [13] Biobase_2.67.0 GenomicRanges_1.57.2
## [15] GenomeInfoDb_1.41.2 IRanges_2.39.2
## [17] S4Vectors_0.43.2 BiocGenerics_0.53.0
## [19] MatrixGenerics_1.17.1 matrixStats_1.4.1
## [21] EBImage_4.47.1 BiocStyle_2.35.0
##
## loaded via a namespace (and not attached):
## [1] bitops_1.0-9 gridExtra_2.3 rlang_1.1.4
## [4] magrittr_2.0.3 svgPanZoom_0.3.4 shinydashboard_0.7.2
## [7] compiler_4.4.1 png_0.1-8 systemfonts_1.1.0
## [10] fftwtools_0.9-11 vctrs_0.6.5 pkgconfig_2.0.3
## [13] SpatialExperiment_1.15.1 crayon_1.5.3 fastmap_1.2.0
## [16] magick_2.8.5 XVector_0.45.0 labeling_0.4.3
## [19] utf8_1.2.4 promises_1.3.0 rmarkdown_2.28
## [22] UCSC.utils_1.1.0 ggbeeswarm_0.7.2 xfun_0.48
## [25] zlibbioc_1.51.2 cachem_1.1.0 jsonlite_1.8.9
## [28] highr_0.11 later_1.3.2 rhdf5filters_1.17.0
## [31] Rhdf5lib_1.27.0 BiocParallel_1.39.0 jpeg_0.1-10
## [34] tiff_0.1-12 terra_1.7-83 parallel_4.4.1
## [37] R6_2.5.1 bslib_0.8.0 RColorBrewer_1.1-3
## [40] jquerylib_0.1.4 Rcpp_1.0.13 knitr_1.48
## [43] httpuv_1.6.15 nnls_1.6 tidyselect_1.2.1
## [46] yaml_2.3.10 viridis_0.6.5 codetools_0.2-20
## [49] lattice_0.22-6 tibble_3.2.1 withr_3.0.2
## [52] shiny_1.9.1 evaluate_1.0.1 pillar_1.9.0
## [55] BiocManager_1.30.25 generics_0.1.3 sp_2.1-4
## [58] RCurl_1.98-1.16 munsell_0.5.1 scales_1.3.0
## [61] xtable_1.8-4 glue_1.8.0 maketools_1.3.1
## [64] tools_4.4.1 sys_3.4.3 locfit_1.5-9.10
## [67] buildtools_1.0.0 grid_4.4.1 colorspace_2.1-1
## [70] GenomeInfoDbData_1.2.13 raster_3.6-30 beeswarm_0.4.0
## [73] vipor_0.4.7 cli_3.6.3 fansi_1.0.6
## [76] viridisLite_0.4.2 svglite_2.1.3 dplyr_1.1.4
## [79] gtable_0.3.6 sass_0.4.9 digest_0.6.37
## [82] farver_2.1.2 rjson_0.2.23 htmlwidgets_1.6.4
## [85] htmltools_0.5.8.1 lifecycle_1.0.4 httr_1.4.7
## [88] mime_0.12