Understanding and controlling the effect of lossy raw data compression on CT images

Department of Electrical Engineering, Stanford University, Stanford, California 94305, USA.
Medical Physics (Impact Factor: 3.01). 08/2009; 36(8):3643-53. DOI: 10.1118/1.3158738
Source: PubMed

ABSTRACT The requirements for raw data transmission through a CT scanner slip ring, through the computation system, and for storage of raw CT data can be quite challenging as scanners continue to increase in speed and to collect more data per rotation. Although lossy compression greatly mitigates this problem, users must be cautious about how errors introduced manifest themselves in the reconstructed images. This paper describes two simple yet effective methods for controlling the effect of errors in raw data compression and describe the impact of each stage on the image errors. A CT system simulator (CATSIM, GE Global Research Center, Niskayuna, NY) was used to generate raw CT datasets that simulate different regions of human anatomy. The raw data are digitized by a 20-bit ADC and companded by a log compander. Lossy compression is performed by quantization and is followed by JPEG-LS (lossless), which takes advantage of the correlations between neighboring measurements in the sinogram. Error feedback, a previously proposed method that controls the spatial distribution of reconstructed image errors, and projection filtering, a newly proposed method that takes advantage of the filtered backprojection reconstruction process, are applied independently (and combined) to study their intended impact on the control and behavior of the additional noise due to the compression methods used. The log compander and the projection filtering method considerably reduce image error levels, while error feedback pushes image errors toward the periphery of the field of view. The results for the images are a compression ratio (CR) of 3 that keeps peak compression errors under 1 HU and a CR of 9 that increases image noise by only 1 HU in common CT applications. Lossy compression can substantially reduce raw CT data size at low computational cost. The proposed methods have the flexibility to operate at a wide range of compression ratios and produce predictable, object-independent, and often imperceptible image artifacts.

1 Follower
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In fields ranging from radio astronomy to magnetic resonance imaging, Fourier inversion of data not falling on a Cartesian grid has been a problem. As a result, multiple algorithms have been created for reconstructing images from non-uniform frequency samples. In the technique known as gridding, the data samples are weighted for sampling density and convolved with a finite kernel, then resampled on a grid preparatory to a fast Fourier transform. This paper compares the utility of several convolution functions, including one that out-performs the "optimal" prolate spheroidal wave function in some situations.
    IEEE Transactions on Medical Imaging 10/1991; 10(3-10):473 - 478. DOI:10.1109/42.97598 · 3.80 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: We present a theoretical overview and a performance evaluation of a novel z-sampling technique for multidetector row CT (MDCT), relying on a periodic motion of the focal spot in the longitudinal direction (z-flying focal spot) to double the number of simultaneously acquired slices. The z-flying focal spot technique has been implemented in a recently introduced MDCT scanner. Using 32 x 0.6 mm collimation, this scanner acquires 64 overlapping 0.6 mm slices per rotation in its spiral (helical) mode of operation, with the goal of improved longitudinal resolution and reduction of spiral artifacts. The longitudinal sampling distance at isocenter is 0.3 mm. We discuss in detail the impact of the z-flying focal spot technique on image reconstruction. We present measurements of spiral slice sensitivity profiles (SSPs) and of longitudinal resolution, both in the isocenter and off-center. We evaluate the pitch dependence of the image noise measured in a centered 20 cm water phantom. To investigate spiral image quality we present images of an anthropomorphic thorax phantom and patient scans. The full width at half maximum (FWHM) of the spiral SSPs shows only minor variations as a function of the pitch, measured values differ by less than 0.15 mm from the nominal values 0.6, 0.75, 1, 1.5, and 2 mm. The measured FWHM of the smallest slice ranges between 0.66 and 0.68 mm at isocenter, except for pitch 0.55 (0.72 mm). In a centered z-resolution phantom, bar patterns up to 15 lp/cm can be visualized independent of the pitch, corresponding to 0.33 mm longitudinal resolution. 100 mm off-center, bar patterns up to 14 lp/cm are visible, corresponding to an object size of 0.36 mm that can be resolved in the z direction. Image noise for constant effective mAs is almost independent of the pitch. Measured values show a variation of less than 7% as a function of the pitch, which demonstrates correct utilization of the applied radiation dose at any pitch. The product of image noise and square root of the slice width (FWHM of the respective SSP) is the same constant for all slices except 0.6 mm. For the thinnest slice, relative image noise is increased by 17%. Spiral windmill-type artifacts are effectively suppressed with the z-flying focal spot technique, which has the potential to maintain a low artifact level up to pitch 1.5, in this way increasing the maximum volume coverage speed that can be clinically used.
    Medical Physics 09/2005; 32(8):2536-47. DOI:10.1118/1.1949787 · 3.01 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Proprietary compression schemes have a cost and risk associated with their support, end of life and interoperability. Standards reduce this cost and risk. The new JPEG-LS process (ISO/IEC 14495-1), and the lossless mode of the proposed JPEG 2000 scheme (ISO/IEC CD15444-1), new standard schemes that may be incorporated into DICOM, are evaluated here. Three thousand, six hundred and seventy-nine (3,679) single frame grayscale images from multiple anatomical regions, modalities and vendors, were tested. For all images combined JPEG-LS and JPEG 2000 performed equally well (3.81), almost as well as CALIC (3.91), a complex predictive scheme used only as a benchmark. Both out-performed existing JPEG (3.04 with optimum predictor choice per image, 2.79 for previous pixel prediction as most commonly used in DICOM). Text dictionary schemes performed poorly (gzip 2.38), as did image dictionary schemes without statistical modeling (PNG 2.76). Proprietary transform based schemes did not perform as...
    Proceedings of SPIE - The International Society for Optical Engineering 05/2000; DOI:10.1117/12.386389 · 0.20 Impact Factor
Show more

Adam S Wang