Medical imaging equipment generates prodigious amounts of high-speed sampled data. Computed tomography (CT) scanners are no different, as they irradiate patients with a thin wedge of x-rays, and the x-ray source rotates around the patient.
Attenuated x-rays are captured by multiple analog-to-digital converters (ADCs) (Fig. 1) that operate up to 10 ksamples/s at 20 bits/sample. CT x-ray detectors are built in a 2D matrix with about 1000 sensors per “slice” (one row of detectors). The number of slices varies between 16 slices in a low-end CT scanner and 320 slices in a high-end CT scanner. Thus, a high-end CT scanner generates up to 64 Gbits/s of real-time data.
This sampled x-ray data must be sent across the CT scanner’s rotating interface (called a slip ring) before being converted into 3D patient volumes by CT image reconstruction algorithms. Because these software reconstruction algorithms take five to 15 minutes to convert 30 seconds of x-ray samples into the 3D volumes that radiologists diagnose, the raw samples are first captured by high-speed disk drive arrays.
The CT scanner’s slip ring and disk drive arrays together can account for 20% of a CT scanner’s bill of materials cost. Compression reduces the cost of CT scanners by enabling lower-bandwidth slip rings and lower-cost storage arrays, with no apparent impact on image quality.
A recent conference paper demonstrated that raw x-ray samples can be compressed by 4:1 without affecting clinical image quality.1 In this paper, a Stanford radiologist compared 10 side-by-side image volumes (over 400 image pairs) and tried to identify which images were generated from non-compressed versus compressed x-ray samples (Fig 2).
While a few of the images had detectable artifacts, the radiologist concluded that his clinical diagnoses in all cases would have been identical using the images created from 4x-compressed CT x-ray data. The Prism CT compression algorithm used to compress the x-ray samples achieves about 2:1 lossless compression by exploiting sample-to-sample and row-to-row correlation.
To achieve 4:1 compression, Prism CT varies quantization parameters across rows and columns, since not all x-ray samples are equally important. As one might expect, x-ray samples that do not traverse the patient do not contribute to the diagnostic patient image. One can safely under-sample these out-of-body x-rays without affecting clinical image quality.
Using 4:1 compression, CT scanner companies can compress x-ray samples during disk writes and decompress them during disk reads prior to image reconstruction. By adding compression transparently to the CT storage array itself, CT vendors can easily integrate compressing storage arrays that cost significantly less than arrays that do not use compression.
For example, hard-disk drives for CT storage arrays sustain write speeds of 100 Mbytes/s, so CT manufacturers require 80 such disk drives to capture 64 Gbits/s (8 Gbytes/s). Even at a low price of $100 per drive, the drives themselves account for $8000 of CT cost. Adding drive controllers and redundancy adds thousands of dollars more to CT storage array cost.
So with a 4x savings in the CT storage array, one of its most expensive components, CT scanner prices can be reduced. And that will eventually translate into lower health care costs for us all.
1. Wegener, A., Senzig, R., Chandra, N., Ling, Y., Herfkens, R., “Effects of fixed-rate CT projection data compression on perceived and measured CT image quality,” SPIE Medical Imaging Conference, Vol. 7627-15 (Feb. 2010); spie.org/x648.html?product_id=841145