Properties generated by OpenSlide

openslide.background-color

openslide.bounds-height

openslide.bounds-width

openslide.bounds-x

openslide.bounds-y

openslide.comment

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

openslide.quickhash-1

openslide.vendor

Properties for TIFF-based formats

tiff.Artist

tiff.Copyright

tiff.DateTime

tiff.DocumentName

tiff.HostComputer

tiff.ImageDescription

tiff.Make

tiff.Model

tiff.ResolutionUnit

tiff.Software

tiff.XPosition

tiff.XResolution

tiff.YPosition

tiff.YResolution

Vendor-specific properties

A list of vendor-specific properties can be found on the pages for each vendor format[1].

Vendor Documentation

http://www.aperio.com/documents/api/Aperio_Digital_Slides_and_Third-party_data_interchange.pdf

Detection

Aperio slides are stored in single-file TIFF format. OpenSlide will detect a file as Aperio if:

Relevant TIFF tags

Extra data stored in ImageDescription

For tiled images, the ImageDescription tag contains some dimensional downsample information as well as what look like offsets. Additionally, vertical line-delimited key-value pairs are stored, in at least the full-resolution image. A key-value pair is equals-delimited. These key-values are stored as properties starting with “aperio.”. Currently, OpenSlide does not use any of the information present in these key-value fields.

For stripped images, the ImageDescription tag may contain a name, followed by a carriage return. This is used for naming the associated images. The second image in the file does not have a name, though it is an associated image.

TIFF Image Directory Organization

http://www.aperio.com/documents/api/Aperio_Digital_Slides_and_Third-party_data_interchange.pdf page 14:

The first image in an SVS file is always the baseline image (full resolution). This image is always tiled, usually with a tile size of 240x240 pixels. The second image is always a thumbnail, typically with dimensions of about 1024x768 pixels. Unlike the other slide images, the thumbnail image is always stripped. Following the thumbnail there may be one or more intermediate “pyramid” images. These are always compressed with the same type of compression as the baseline image, and have a tiled organization with the same tile size.

Optionally at the end of an SVS file there may be a slide label image, which is a low resolution picture taken of the slide’s label, and/or a macro camera image, which is a low resolution picture taken of the entire slide. The label and macro images are always stripped.

JPEG 2000 (compression types 33003 or 33005)

Some Aperio files use compression type 33003 or 33005. Images using this compression need to be decoded as a JPEG 2000 codestream. For 33003: YCbCr format, possibly with a chroma subsampling of 4:2:2. For 33005: RGB format. Note that the TIFF file may not encode the colorspace or subsampling parameters in the PhotometricInterpretation field, nor the YCbCrSubsampling field, even though the TIFF standard seems to require this. The correct subsampling can be found in the JPEG 2000 codestream.

Associated Images

thumbnail

label

macro

Known Properties

All key-value data encoded in the ImageDescription TIFF field is represented as properties prefixed with “aperio.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Aperio/

Detection

OpenSlide will detect a file as Hamamatsu if:

or if:

Overview

The Hamamatsu format has three variants. VMS and VMU consist of an index file, 2 or more image files, and (in the case of VMS) an “optimisation” file. NDPI consists of a single TIFF-like file with some custom TIFF tags. VMS and NDPI contain JPEG images; VMU contains NGR images (a custom uncompressed format).

Multiple focal planes are ignored, only focal plane 0 is read.

JPEG does not allow for files larger than 65535 pixels on a side. In VMS, multiple JPEG files are used to encode large images. To avoid having many files, VMS uses close to maximum size (65K by 65K) JPEG files. NDPI, instead, stuffs large levels into a single JPEG and sets the overflowed width/height fields to 0.

Unfortunately, JPEG provides very poor support for random-access decoding of parts of a file. To get around this, JPEG restart markers are placed at regular intervals, and these offsets are specified in the optimisation file (in VMS) or in a TIFF tag (in NDPI). With restart markers identified, OpenSlide can treat JPEG as a tiled format, where the height is the height of an MCU row, and the width is the number of MCUs per row divided by the restart marker interval times the width of an MCU. (This often leads to oddly-shaped and inefficient tiles of 4096x8, for example.)

Unfortunately, the VMS optimisation file does not give the location of every restart marker, only the ones found at the beginning of an MCU row. It also seems that the file ends early, and does not give the location of the restart marker at the last MCU row of the last image file.

Thus, the optimisation file can only be taken as a hint, and cannot be trusted. The entire set of JPEG files must be scanned for restart markers in order to facilitate random access. OpenSlide does this lazily as needed, and also in a background thread that runs only when OpenSlide is otherwise idle.

The VMS map file is a lower-resolution version of the other images, and can be used to make a 2-level JPEG pyramid. JPEG also allows for lower-resolution decoding, so further pyramid levels are synthesized from each JPEG file.

VMS File

The .vms file is the main index file for the VMS format. It is a Windows INI-style key-value pair file, with sections. Only keys in the Virtual Microscope Specimen group are read by OpenSlide.

Here are known keys from the file:

VMU File

The .vmu file is the main index file for the VMU format. Only keys in the Uncompressed Virtual Microscope Specimen group are read by OpenSlide.

Here are known keys from the file:

NDPI File

NDPI uses a TIFF-like structure, but libtiff cannot read the headers of an NDPI file. This is because NDPI specifies the RowsPerStrip as the height of the file, and after doing out the multiplication, this typically overflows libtiff and it refuses to open the file. Also, the TIFF tags are not stored in sorted order.

NDPI stores an image pyramid in TIFF directory entries. In some files, the lower-resolution pyramid levels contain no restart markers. The macro image, and sometimes an active-region map, seems to come last.

JPEG files in NDPI are not necessarily valid. If ImageWidth or ImageHeight exceeds the JPEG limit of 65535, then the width or height as stored in the JPEG file is 0. libjpeg will refuse to read the header of such a file, so the JPEG data stream must be altered when fed into libjpeg.

NDPI is based on the classic TIFF format, which does not support files larger than 4 GB. However, NDPI files can be larger than 4 GB. NDPI generally handles this by overflowing the corresponding TIFF fields, requiring the reader to guess the high-order bits. This affects TIFF Value Offsets with pointers to out-of-line values, as well as the value of the StripOffsets field. Some TIFF fields (e.g. Software) have the same Value Offset in every directory; for these, no concatenation of high-order bits is necessary. For the others (primarily field 65426) it seems reasonable to select high-order bits which place the value at the largest offset below the directory itself, since the TIFF directory is positioned after the data it points to. NDPI always stores next-directory offsets (in the TIFF header and at the end of each directory) as 64-bit quantities, even though TIFF specifies them as 32 bits; this is possible because the TIFF standard places them at the end of their parent data structures.

It is not clear whether NDPI can support individual directories larger than 4 GB. Such files would require additional inferences for the StripByteCounts field, for Value Offsets that are identical across directories, and for the optimisation entries.

Here are the observed TIFF tags:

Optimisation File (only for VMS)

The optimisation file contains a list of 32- (or 64- or 320- ?) bit little endian values, giving the file offset into an MCU row, each offset starts at a 40-byte alignment, and the last row (of the entire file, not each image) seems to be missing. The offsets are all packed into 1 file, even with multiple images. The order of images is left-to-right, top-to-bottom.

Map File (only for VMS/VMU)

The VMS map file is a standard JPEG file. Its restart markers (if any) are not included in the optimisation file. The VMU map file is in NGR format. This file can be used to provide a lower-resolution view of the slide.

Image Files (only for VMS/VMU)

These files are given by the VMS/VMU ImageFile keys. They are assumed to have a height which is a multiple of the MCU height. They are assumed to have a width which is a multiple of MCUs per row divided by the restart interval.

For VMS, these files are in JPEG, for VMU they are in NGR format.

NGR Format

The NGR file contains uncompressed 16-bit RGB data, with a small header. The files we have encountered start with GN, two more bytes, and then width, height, and column width in little endian 32-bit format. The column width must divide evenly into the width. Column width is important, since NGR files are generated in columns, where the first column comes first in the file, followed by subsequent files. Columns are painted left-to-right.

At offset 24 is another 32-bit integer which gives the offset in the file to the start of the image data. The image data we have encountered is in 16-bit little endian format.

Associated Images

macro

Known Properties

All key-value data stored in the VMS/VMU file, and known tags from the NDPI file, are encoded as properties prefixed with “hamamatsu.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

NDPI format

VMS format

Detection

Leica slides are stored in single-file BigTIFF format. OpenSlide will detect a file as Leica if:

To open Leica files, OpenSlide must be built with libtiff 4 or above.

Relevant TIFF tags

File Organization

The ImageDescription tag of the first TIFF directory contains an XML document that defines the structure of the slide.

Leica slides are structured as a collection of images, each of which has multiple dimensions (pyramid levels). The collection has a size, and images have a size and position, measured in nanometers. Each dimension has a size in pixels, an optional focal plane number, and a TIFF directory containing the image data. Fluorescence images have different dimensions (and thus different TIFF directories) for each channel. OpenSlide currently rejects fluorescence images and ignores focal planes other than plane 0.

Brightfield slides have at least two images: a low-resolution macro image and one or more main images corresponding to regions of the macro image. The macro image has a position of (0, 0) and a size matching the size of the collection. Fluorescence slides can have two macro images: one brightfield and one fluorescence.

The slide provides enough information to composite the various images, including the macro image, into a single pyramid. However, there are some complications: <listitem>The resolution of the macro image is generally not related to the resolution of the main images by a power of two. </listitem><listitem>Downsampled dimensions are generally downsampled from the next larger dimension by a factor of 4, but main images can be scanned with distinct objectives that may differ by only a factor of 2. </listitem>.PP Thus, in general, the images in a collection cannot be rendered into a unified pyramid without scaling the original pixel data. OpenSlide does not attempt to do this. Instead, OpenSlide omits the macro image from the pyramid and refuses to open slides whose main images have inconsistent resolutions.

Associated Images

macro

Known Properties

leica.aperture

leica.barcode

leica.creation-date

leica.device-model

leica.device-version

leica.illumination-source

leica.objective

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Leica/

Detection

OpenSlide will detect a file as MIRAX if:

Overview

MIRAX can store slides in JPEG, PNG, or BMP formats. Because JPEG does not allow for large images, and JPEG and PNG provide very poor support for random-access decoding of part of an image, multiple images are needed to encode a slide. To avoid having many individual files, MIRAX packs these images into a small number of data files. The index file provides offsets into the data files for each required piece of data.

The camera on MIRAX scanners takes overlapping photos and records the position of each one. Each photo is then split into multiple images which do not overlap. Overlaps only occur between images that come from different photos.

To generate level n + 1, each image from level n is downsampled by 2 and then concatenated into a new image, 4 old images per new image (2 x 2). This process is repeated for each level, irrespective of image overlaps. Therefore, at sufficiently high levels, a single image can contain one or more embedded overlaps of non-integral width.

Index File

The index file starts with a five-character ASCII version string, followed by the SLIDE_ID from the slidedat file. The rest of the file consists of 32-bit little-endian integers (unaligned), which can be data values or pointers to byte offsets within the index file.

The first two integers point to offset tables for the hierarchical and nonhierarchical roots, respectively. These tables contain one record for each VAL in the HIERARCHICAL slidedat section. For example, the record for NONHIER_1_VAL_2 would be stored at nonhier_root + 4 * (NONHIER_0_COUNT + 2).

Each record is a pointer to a linked list of data pages. The first two values in a data page are the number of data items in the page and a pointer to the next page. The first page always has 0 data items, and the last page has a 0 next pointer.

There is one hierarchical record for each zoom level. The record contains data items consisting of an image index, offset and length within a file, and a file number. The file number can be converted to a data file name via the DATAFILE slidedat section. The image index is equal to image_y * GENERAL.IMAGENUMBER_X + image_x. Image coordinates which are not multiples of the zoom level’s downsample factor are omitted.

Nonhierarchical records refer to associated images and additional metadata. Nonhierarchical data items consist of three zero values followed by an offset, length, and file number as in hierarchical records.

Data Files

A data file begins with a header containing a five-character ASCII version string, the SLIDE_ID from the slidedat file, the file number encoded into three ASCII characters, and 256 bytes of padding. (In newer slides, the SLIDE_ID and file number are encoded as UTF-16LE, so the second half of each value is truncated away.) The remainder of the file contains packed data referenced by the index file.

Slide Position File

The slide position file is referenced by the VIMSLIDE_POSITION_BUFFER.default nonhierarchical section. It contains one entry for each camera position (not each image position) in row-major order. Each entry is nine bytes: a flag byte, the X pixel coordinate of the photo (4 bytes, little-endian, may be negative), and the Y coordinate (4 bytes, little-endian, may be negative). In slides with CURRENT_SLIDE_VERSION ≥ 1.9, the flag byte is 1 if the slide file contains images for this camera position, 0 otherwise. In older slides, the flag byte is always 0.

In slides with CURRENT_SLIDE_VERSION ≥ 2.2, the slide position file is compressed with DEFLATE and referenced by the StitchingIntensityLayer.StitchingIntensityLevel nonhierarchical section.

Associated Images

thumbnail

label

macro

Known Properties

All key-value data stored in the Slidedat.ini file are encoded as properties prefixed with “mirax.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

See Also

Introduction to MIRAX/MRXS[3]. Note that our terminology has changed since that document was written; where it says “tile”, substitute “image”, and where it says “subtile”, substitute “tile”.

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Mirax/

Detection

Philips TIFF files are stored in single-file TIFF or BigTIFF format. OpenSlide will detect a file as Philips if:

To open BigTIFF files, OpenSlide must be built with libtiff 4 or above.

File Organization

Philips TIFF is an export format. The native Philips format, iSyntax, is a custom multi-file format not currently supported by OpenSlide.

The ImageDescription tag of the first TIFF directory contains an XML document with a hierarchical structure containing key-value pairs. The keys are based on DICOM tags.

The level dimensions given in the TIFF ImageWidth and ImageLength fields, and also in the ImageDescription XML, are merely the TIFF tile size multiplied by the number of tiles in each dimension. Thus, they include the size of the padding in the right-most column and bottom-most row of tiles. Each level typically uses the same tile size but requires a different amount of padding, so the aspect ratios of the levels are inconsistent and the level dimensions are not proportional to the level downsamples. Correct downsamples can be calculated from the levels’ pixel spacings in the XML metadata.

Slides with multiple regions of interest are structured as a single image pyramid enclosing all regions. Slides may omit pixel data for TIFF tiles not in an ROI; this is represented as a TileOffset of 0 and a TileByteCount of 0. When such tiles are downsampled into a tile that does contain pixel data, their contents are rendered as white pixels.

Label and macro images are stored as Base64-encoded JPEGs in the ImageDescription XML. Some slides also store these images as stripped TIFF directories whose ImageDescriptions start with Label and Macro, respectively.

Relevant TIFF tags

Associated Images

label

macro

Known Properties

All key-value data encoded in the DPUfsImport object, in the first DPScannedImage object with a PIM_DP_IMAGE_TYPE of WSI, and in that object’s PixelDataRepresentation objects is represented as properties prefixed with “philips.”.

openslide.mpp-x

openslide.mpp-y

Test Data

No public data available. Contact the mailing list[4] if you have some.

Detection

OpenSlide will detect a file as Sakura if:

File Organization

Sakura slides are SQLite 3 database files written by the eXpress Persistent Objects ORM. Tables contain slide metadata, associated images, and JPEG tiles. Tiles are addressed as (focal plane, downsample, level-0 X coordinate, level-0 Y coordinate, color channel), with separate grayscale JPEGs for each color channel. Despite the generality of the address format, tiles appear to be organized in a regular grid, with power-of-two level downsamples and without overlapping tiles. The structure of the file allows scans to be sparse, but it is not clear if this is actually done.

SQL Tables

Some irrelevant tables and columns have been omitted from the summary below. DataManagerSQLiteConfigXPO.PP Useful only to get a reference to the unique table. OpenSlide requires this table to contain exactly one row.

SVSlideDataXPO.PP High-level metadata about a slide. OpenSlide assumes this table will contain exactly one row.

SVHRScanDataXPO.PP A single high-resolution scan of a slide from SVSlideDataXPO. OpenSlide assumes this table will contain exactly one row.

SVScannedImageDataXPO.PP Contains associated images other than the thumbnail.

tile.PP This table is most naturally used to map tile coordinates to tile IDs, but is not suitable for individual lookups because it has no useful indexes. In addition, some Sakura slides don’t have it. OpenSlide ignores the table and constructs tile IDs directly from tile coordinates.

Unique table

This is the table named by DataManagerSQLiteConfigXPO.TableName. It contains named blobs including the JPEG tile data.

This table stores a variety of blob types.

Header blob

The Header blob is a small binary structure containing little-endian integers as follows:

Associated Images

label

macro

thumbnail

Known Properties

sakura.Creator

sakura.Date

sakura.Description

sakura.DiagnosisCode

sakura.FocussingMethod

sakura.Keywords

sakura.NominalLensMagnification

sakura.ResolutionMmPerPix

sakura.ScanId

sakura.SlideId

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

No public data available. Contact the mailing list[4] if you have some.

Detection

Trestle slides are stored in single-file TIFF format. OpenSlide will detect a file as Trestle if:

Relevant TIFF tags

Extra data stored in ImageDescription

The ImageDescription tag contains semicolon-delimited key-value pairs. A key-value pair is equals-delimited. We use the OverlapsXY and Background Color keys from the ImageDescription, and ignore the rest. All of these values are stored as properties starting with “trestle.”.

TIFF Image Directory Organization

The first image in the TIFF file is the full-resolution image. The subsequent images are assumed to be decreasingly sized reduced-resolution images.

Overlaps

The OverlapsXY pseudo-field encodes a list of tile overlap values as ASCII.

Example: “64 64 32 32 16 16” (note the initial space).

These values represent the standard overlaps between adjacent tiles in X and Y, in pixels. This example encodes 3 levels worth of overlaps. Further overlaps are assumed to have the value 0.

Individual tile overlaps may differ from the standard overlaps. These individual overlaps are recorded in .tif-Nb files adjacent to the .tif file, where N is the level number. OpenSlide does not read these files, though they have been partially decoded; see issue 21[5] for details.

Associated Images

macro

Known Properties

All data encoded in the ImageDescription TIFF field is represented as properties prefixed with “trestle.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Trestle/

Detection

Ventana slides are stored in single-file BigTIFF format. OpenSlide will detect a file as Ventana if:

To open Ventana files, OpenSlide must be built with libtiff 4 or above.

Associated Images

macro

thumbnail

Known Properties

All XML attributes in the iScan element are represented as properties prefixed with “ventana.”.

openslide.mpp-x

openslide.mpp-y

openslide.objective-power

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Ventana/

Detection

OpenSlide will detect a file as generic TIFF if:

TIFF Image Directory Organization

The first image in the TIFF file is the full-resolution image. Any other tiled images in the file with the “reduced resolution” bit set are assumed to be reduced-resolution versions of the original.

Associated Images

None.

Known Properties

Many TIFF tags are encoded as properties starting with “tiff.”.

Test Data

http://openslide.cs.cmu.edu/download/openslide-testdata/Generic-TIFF/