Revised 2001/05/30 dclunie@dclunie.com Replaced DICOMDIR with version that has correct High Bit value in Icon Image Sequence (was 8 should have been 7). ----- This disk consists of a collection of medical images for the purpose of informally evaluating the effect of proposed JPEG 2000 schemes on image quality. They contain no patient identifying information. They have been obtained from various publically available collections of images, including NEMA DICOM demonstrations and public web servers. However, copyright is retained by the owners of the original images and hence they should not be distributed beyond the intended purpose of the JPEG evaluation. They are intended to be a representative set of the broad range of different types of medical images. They may not necessarily particularly challenging or unusual and may not contain any particularly subtle abnormalities, so should not be assumed to be critical tests of a compression scheme nor imply any likely usefulness from a diagnostic point of view. The decision to use a lossy compression scheme for medical applications may require a more rigorous observer-performance study of a carefully selected set to test a particular diagnostic task. The images are "dual-personality" in the sense that they are both DICOM and TIFF images. The collection includes a dicom directory and is written with software conforming to the General Purpose CD-R Media Application Profile. Images are all encoded as current standard DICOM Image Storage SOP Classes - the Visible Light images are stored as Secondary capture, not the proposed VL Image Storage SOP Class, which may not yet be supported on the review workstations used for the evaluation. Non-medical users may find it more convenient to extract the pixel data from the images using a conventional TIFF reader, rather than a DICOM reader. Be aware that some of the images, including the CT and some of the MR images, contained signed data, and hence will "look strange". It is expected that the results of the compress/decompress cycle will result in data of the same form as the original and will be evaluated visually using medical software that "knows" about signed data, ie. don't try to fix what these "look like" before compressing them ! The following images are present: Image File Pixel Data Rows Cols Bits Samples Zero Order Bytes Offset Entropy CT Lung Contrast CT1 524288 6176 512 512 16 1 8.20914 CT Brain Contrast CT2 524288 1426 512 512 16 1 6.8012 Mammogram MG1 28580992 1186 4664 3064 12 1 10.561 MR Posterior Fossa MR1 524288 1430 512 512 16 1 10.083 MR Shoulder MR2 2097152 1728 1024 1024 12 1 6.89806 MR Knee MR3 524288 9272 512 512 16 1 6.30807 MR Brain MR4 524288 1748 512 512 12 1 6.03216 NM Bonescan NM1 524288 2632 1024 256 16 1 3.99372 Chest X-Ray RG1 7198310 1570 1955 1841 15 1 13.7073 Hip Joint RG2 7532800 1134 2140 1760 10 1 7.20069 Bone Lesion Leg RG3 6195200 1202 1760 1760 10 1 6.62469 US Color Tissue Doppler US1 921600 1044 480 640 8 3 3.11635 XA Neuroangiogram XA1 2097152 936 1024 1024 10 1 6.53488 Scanned Film CT Brain SC1 10186752 880 2487 2048 12 1 5.75373 GI Endoscopy Esophagus VL1 1102248 894 486 756 8 3 2.66609 GI Endoscopy Stomach VL2 1102248 890 486 756 8 3 2.99886 GI Endoscopy Colon VL3 1102248 900 486 756 8 3 2.55583 Retinal Fundoscopy VL4 12474504 870 1868 2226 8 3 4.32233 Pathology Whole Prostate VL5 26753400 886 3340 2670 8 3 6.16013 Pathology Stomach Biopsy VL6 1102248 902 486 756 8 3 4.53285 Note that Bits is the official number of bits used as stated by the source of the images ... the dynamic range of the actual image may not use this much. Also, the images are stored as 8 or 16 bit words (little endian), eg. a 12 bit pixel actually occupies 16 bits (in the least significant bits). This is all described in the DICOM header, but the TIFF header doesn't preserve this distinction. All images are single frame. The color image(s) are all color-by pixel (ie. RGBRGB...) not color-by-plane. david.clunie@med.ge.com 1998/03/09