DICOM Converted and Original MRXS Whole Slide Images of Paired Hematoxylin-Eosin +/- Saffron Canine Breast Cancer Specimens

Overview. This dataset contains DICOM converted and original MRXS Whole Slide Images of canine breast cancer specimens with paired Hematoxylin-Eosin with and without Saffron staining, described in the scientific paper [1]. The source dataset comprises seven canine breast cancer specimens prepared as HE slides then additionally stained with Saffron. Both versions of the slides were scanned at 20x (approx. 0.0005mm per pixel) with a 3DHistech Panoramic Scan II, as described in the original paper [1]. The source images were shared in the proprietary MRXS form, and then converted to DICOM using the 3DHISTECH SlideMaster [2] software (v3.1.0.254167) with a temporary license provided by the vendor. The conversion tool properties were configured to write TILED_FULL DICOM images with Extended Offset Tables. Additional metadata was then added describing the subject (patient) and specimen preparation using the open source com.pixelmed.apps.SetCharacteristicsFromSummary tool [3] with the information derived from the original paper and the source file name, using the Bourne shell script in this collection. Specimens and slides and scans were acquired at Veterinärmedizinische Universität Wien, and are publicly shared (CC-BY-4.0) with the permission of Tribun Health, who sponsored the original study [1]. Notes. Though mathematically lossless conversion of proprietary WSI format to DICOM is often possible by using the original compressed tiles [4], the MRXS files contain overlapping tiles, so the conversion to the conventional non-overlapped DICOM TILED_FULL representation is inherently lossy due to (a) the resolution of overlapping as well as (b) the decompression and recompression cycle required for re-rendering; both of these factors are affected by the specific converter code and settings used; see the discussion in [5]. The source MRXS files are provided in case the recipient wants to directly access the original tiles, or use alternative converters, such as wsidicomizer [6], which supports MRXS and rendering of overlapping tiles through the use of the OpenSlide reader [7]. The LABEL and THUMBNAIL images in the vendor-converted DICOM series appear to contain inconsistencies with respect to the size measurements as reported by dciodvfy [8], so caution should be taken interpreting these values. Acknowledgements. We would like to acknowledge the support of Tribun Health who agreed to publicly share the images, and the slide scanner vendor, 3DHISTECH, who provided a temporary license to support DICOM conversion.