Introduction
Conventional analogue approaches to the storage of angiogram video are now being superseded by their digital counterparts. The advantages of this new approach include the potential for higher quality images, no degradation in quality over time, faster access, improved display management and ease of transmission, to name but a few. These in turn create exciting new possibilities, greatly improving efficiency, while at the same time relaxing geographical constraints on the access of the data allowing a consultant in another hospital, or even another country to review an angiogram sequence remotely.
The Data Rate
Unfortunately, the logistics concerning angiogram video, as with much medical imaging, makes such a vision hard to realise in practice. Typical image sizes are 512x512 (or even 1024x1024) pixels, taken at 30 frames/sec with at least 8 bits of resolution. A typical procedure may be of the order of 5 minutes resulting in approximately 2.5GBytes of raw data. At a constant data rate of 64Kbit/sec it would take 80 hours to transmit this quantity of data, and even at 10Mb/sec it would take 30 minutes.
From this it is clear that a practical system for transmitting angiogram video from one hospital to another requires a high bandwidth connection, but even then a wait of 30 minutes may be considered unacceptable, especially if a semi-real-time diagnosis is required. Matters may be improved considerably by compressing the video stream. Due to the sensitivity of the data however, only lossless or near-lossless algorithms can realistically be used. Unfortunately, the tight constraints imposed by lossless compression usually limit the compression ratio to about 2 or 3:1.
We have two digital cardiac angiographic laboratories performing almost 3000 procedures per annum. The images from each procedure (usually 5 to 10 minutes of video) are stored on individual compact discs (CDs). A back-up record is provided by a VHS video tape recorder which stores consecutive studies on tape.
Each CD must be stored in the hospital. There is a legal requirement for record keeping for 7 years but for most practical purposes ready access to the record must be available for one year. The waiting list for cardiac surgery is up to 9 months and these films must be immediately available when the patient is admitted for surgery (usually the day before the operation). The images must be seen by the cardiac surgeon both when deciding whether to accept the patient for surgery (ie when referred by the cardiologist) and again immediately before surgery to plan the details of the procedure.
There are three visiting cardiologists who use our angiographic facilities, the CDs are taken back to their base hospitals for storage. Referrals for surgery or angioplasty comprise a letter and the CD for the surgeon to view so that in effect these CDs are returned to the QE.
In addition, a large number of cardiologists use their own angiographic laboratories but refer patients both routinely and urgently for cardiac surgery or angioplasty. In these cases angiograms are sent by post or in urgent cases with the patient in the ambulance.
Data Access
The major difficulty is storage and
retrieval. A large amount of
time is spent by many personnel (nurses, doctors, X ray staff, secretaries) in
locating films. VHS video back-up record are often required and operations can be delayed when angiograms are missing. The problem is exacerbated by the long
surgical waiting time and the need for physical transport of angiograms (when
referring patients for surgery or angioplasty) between hospitals and within the
QE between the departments of cardiology and cardiac surgery. With such a large number of angiograms
locating the correct set of images from a large storage facility is not always
easy. Patients are accepted for urgent
angioplasty or surgery from cardiologists in hospitals in the West Midlands
without prior viewing of the angiogram.
If the procedure is found to be not possible or appropriate, expensive
hospital beds are used while alternative arrangements are made. Angiograms could be stored on a central
archive accessible by cardiologists and cardiac surgeons inside and outside the
hospital. The possibility of loss or
damage to angiograms over a period of months or years would be much
reduced. There would be no need to
physically transport angiograms again reducing the possibility of loss or
damage. The paper below
considers lossy approaches, with compression ratios of approximately 10:1.
For diagnostic purposes, it is essential that the compression process causes
no tangible loss of detail and introduces no noticeable artefacts which
could be misinterpreted as being pathological in nature. Little research
has been done specifically regarding the compression of angiogram video.
Unfortunately, given the unique structure of the data, video compression
results relating to more conventional types of video (e.g. for digital television),
for which there has been a great deal of research, do not necessarily apply
here. A COMPARATIVE STUDY OF COMPRESSION
METHODOLOGIES FOR DIGITAL ANGIOGRAM VIDEO Summary
Our paper (1.4MBytes) includes a comparative study of various lossy compression techniques for application
to angiogram video. A discussion of the various benefits of digital acquisition
of such data is given, and the resultant quantity of raw video data is assessed.
The prospect of the realistic use of remote analysis and diagnostic techniques
for this type of medical data is then shown to rely quite heavily on a suitable
compression technique. Traditional wavelet
and discrete cosine transform (DCT) based compression methodologies are
both considered, and their performance over a range of bit rates is measured.
In addition to intra-frame coding techniques, inter-frame approaches are
also analysed. This is based on the motion compensated prediction (MCP)
philosophy adopted by most conventional video coding algorithms. Three different
types of MCP are examined, namely block matching, global motion compensation
using an affine flow model with robust motion estimation, and a 'zero inter-frame
motion' approach. Due to the presence of temporally poorly correlated high-frequency
image data, a low-pass spatial filter was found to be an essential requirement
as part of the predictive loop. Although the net effect of using MCP was
beneficial for all three cases, the improvements were most notable at low
bit rates (e.g. 0.2 bits/pixel). At bit rates where the resultant image
was perceived to be 'diagnostically indistinguishable' from the original,
the benefits of MCP were marginal. For the purpose of providing a baseline comparison, we have also implemented a H.263 encoder for angiogram compression (H.M.Lim, B.Eng project 1999). Original, Reconstructed and Error Images RMS Error
Last updated 19 November 2000