Systematical Parameterization, Storage and Representation of Volumetric DICOM Data for Visualization: 3D Presentation States (3DPR) Dr. M. Alper SELVER Dokuz Eylul University Electrical and Electronics Engineering
Research Group FH-Juelich Medical Informatics Prof. Dr. Oğuz Dicle Dr. Sinem Gezer DEU Radiology Prof. Dr. rer. nat. Dr. Felix Fischer Walter Hillen DEU Electrical & Electronics Engineering Dr. Alper Selver
Ultimate Goal Snapshots Videos Movie files Automatic Rendering + all interactions + possible refinements
Part I: Compression of Segmented Data • Archiving result of a segmentation task allows the representation of the segmented volume at a later time. • The segmented volume can be stored in a binary format, which can be restored by a simple combination of the original data with this binary information. • Since, the sizes of the segmented binary data have high memory requirements; a lossless compression method should be employed for efficient archiving. • this study examines different approaches for compression and their suitability for restoring binary segmented data. • To evaluate the compressive properties, multiple test cases with diverse spatial structures and acquired with different modalities from clinical practice have been used.
Methods: (All 2D except octree) • Losless storage of segmented data slice by slice in uncompressed bitmap (BMP), Portable Bitmap (PBM) • Run-length encoding (RLE): effective when a symbol repeatedly occurs in the data. (Java based Birle). • CCITT T.4/T.6: CCITT T.4 consists of a combination of binary run-length coding and a modified Huffman coding. Its disadvantage in short run lengths is corrected in CCITT T.6 by using a Modified Modified Read (MMR) process. (IrfanView 3.98 CCITT Fax 3 and CCITT Fax 4 were selected, respectively). • JBIG2: (Joint Bi-level Image Processing Group) a standard specially designed for binary. JBIG2 is the current version of the standard established by the ISO (International Organization for Standardization) and is also responsible for the JPEG and JPEG2000 standard. (C++ program, jbig2enc). • JPEG 2000: is used for both lossy and lossless compression. The compression process consists of several steps mainly based on the wavelet transform. (The free command-line program GeoJasper) • ZIP: based on the Deflate algorithm, which is composed of two encoding methods: LZ77 and Huffman. With the LZ77 method identical symbol sequences are determined and coded. Then, the coded symbol sequences are compressed using the Huffman method. (IrfanView: by lossless conversion of the PBM file in the graphic format PNG data, images were automatically compressed using the ZIP method). • LZW: uses a sliding window over the data set. The window consists of two buffers: search and preview. The search buffer contains the last coded symbols and is used as a dictionary for symbols from the preview buffer. The LZW method uses separate windows instead of sliding ones. (IrfanView, TIFF format with LZW). • Octree: carried out in two separate files: The values of the voxels are consecutively written to one file. In a second file, the dimensions of the VOI can be stored. With the help of these two files, it is possible to restore the binarized segmentation result as a volume.
Datasets + Aorta (CT with contrast medium) - 250 slices, slice thickness: 1.5 mm - Segmentation: Connected Threshold - VOI: 139x322x288 voxels + Skull (CT) data set consists - 61 slices – Slice thickness 0.7 mm - Segmentation: Connected Threshold - VOI: 175x214x302 voxels Kidney (CT, MR) (different results for interior kidney) - MR coronal, 72 slices ST: 1.4 mm (smoothing: MR-2) - CT series with 238 slices, ST: 1 mm - Segmentation: Fast Marching - VOI (MR:121x52x205) (CT: 114x101x112) voxels. + Skeleton (CT): ribs + hip - 250 slices, slice thickness: 1.5 mm - Segmentation: Connected Threshold - VOI: 239x146x288 voxels.
Results:
Computational Performance
Conclusion • For compression of binary volume data, the JBIG2 method is well suited; however, it is actually optimized for the data reduction of 2D binary images. • It would be better to use a method that can be applied directly to the volume of data without having to split them beforehand into individual slices. • By this way, correlations in the data may be used not only in two, but also in three dimensions. • This requirement could be achieved by modification of the JBIG2 method as follows: – The high compression factor is partly due to the use of a context- based coding, in which the context of the neighbors of the pixel to be encoded. – In current version of JBIG2, the neighbors to be encoded are always the pixels in the same 2D plane (i.e. slice). – By enlarging the vicinity in 3D (i.e. to the adjacent slices), the JBIG2 method could be used without a change in the actual encoding process with a presumably more efficient compression of the binary volume data.
Part II: Parameterization of Segmentation • Final goal: Use segmentation parameters in teleradiology In this study, the goal is: • not to present yet another liver segmentation algorithm • but to report parametric analysis and detailed evaluation of two widely used 3-D segmentation strategies implemented in Insight Toolkit (i.e. under optimized programming conditions)
Segmentation Process Segmentation Post-processing Connected Threshold Pre-processing Filtering (CoT) - VOI Selection (for surface smoothing) Seed points - Filtering Gaussian Upper-Lower limits Gaussian Median Fast Marching (FM) Median Anisotropic - Seed points Anisotropic - diffusion diffusion Maximum gradient Number of iterations 3D-Visualization Volume measurements 3D-rendering
DATASETS Challanges 24 Datasets Average 90 slices 3.2mm Slice thickness liver composed of atypical liver right kidney liver-heart multiple shape appearance boundary components in 2-D
Connected Threshold (CoT) leakage • a member of region growing (RG) process • Initialize seed points • Determine upper and lower threshold levels ROI from Gaussian image Median k=9 k=5 k=15 anisotropic diffusion
Fast Marching • Active contour based (edge-oriented methods) level set equation velocity in the normal direction FM equation
Java Based Object Oriented Implementation
Application Example (Fast Marching)
Evaluation • Volume • Surface (Symmetric Surface Distance-SSD) Each surface voxel of reference volume (V R ) Closest surface voxel of the segmented volume (V S ) SSD: the set of surface voxels V S are given by S(V S ), to the shortest distance of an arbitrary surface voxel of V R
Evaluation
Results
Recommend
More recommend
