Graefe’s Arch Clin Exp Ophthalmol (2002) 240:955–959
S H O R T C O M M U N I C AT I O N
DOI 10.1007/s00417-002-0523-6
Christopher Pesavento Alon Harris Craig Cole Larry Kagemann
Received: 27 February 2002 Revised: 4 June 2002 Accepted: 20 June 2002 Published online: 12 October 2002 © Springer-Verlag 2002
C. Pesavento · A. Harris (✉) · C. Cole L. Kagemann Department of Ophthalmology, Indiana University School of Medicine, 702 Rotary Circle, Indianapolis, IN 46202, USA e-mail:
[email protected] Tel.: +1-317-2740134 Fax: +1-317-2781007 A. Harris Department of Physiology and Biophysics, Indiana University School of Medicine, Indianapolis, Indiana, USA
Improving the analysis of arteriovenous passage times
Abstract Purpose: This study evaluated the effect of two techniques, time-based correction (TBC) and de-interlacing, on the quality of an image sequence from a scanning laser ophthalmoscope (SLO), and their impact on subsequent analysis for the determination of arteriovenous passage (AVP) times. Methods: Fluorescein angiograms obtained from one patient participating in a concurrent study approved by an institutional review board (IRB) were analyzed before and after the serial connection of a TBC, and before and after de-interlacing of the images. Informed consent was obtained, and all tenets of the Declara-
Introduction The development of the scanning laser ophthalmoscope (SLO) during the 1980s introduced a valuable tool with many applications for ocular study and research [5, 11, 12]. One of its most beneficial uses has been in the study of ocular hemodynamics. The SLO has advanced quantitative ocular angiographic studies to new levels, in parameters such as arteriovenous passage (AVP) time as well as choroidal filling time and retinal capillary velocities [1, 2, 13]. In the past ten years, much research has been done and a considerable amount of information gained about ocular hemodynamics, in dozens of studies performed across the world using the SLO. Despite this progress, limits in the technology are still encountered when analyzing angiograms. One such
tion of Helsinki were followed. AVP times were determined and the variance in time per 100 frames was calculated. Results: The average error in the determination of AVP times was significantly less after the application of these techniques (p<0.05). Also, the coefficient of variance in the time elapsed per 100 frames was reduced after TBC and de-interlacing of the video sequence. Conclusion: The study showed that in addition to the subjective improvement in image quality after de-interlacing and TBC, quantitative parameters are improved. This leads to a more accurate analysis of the angiogram.
problem is that the S-video VCR records the angiograms in an interlaced fashion. Interlaced recording systems scan alternate lines to produce two separate fields, one composed of odd lines and one of even lines, which are then “interlaced” together to obtain the full frame of video information. This means that the two fields are separated in time by a very small amount, and motion blurring is introduced in rapidly moving objects [3, 10]. Cardiologists encountered the same problem with the development of percutaneous transluminal coronary angioplasty (PTCA) [3, 8, 10, 14]. Other problems derive from the stretching and wear and tear that videotapes go through in everyday use, which can distort and even break up the picture [9]. In this study, we quantified the effect of two techniques, one hardware and one software, on the quality of SLO video sequences for analysis.
956
Fig. 1 Beginning with line 0, the even lines were obtained from the image of a letter A, and the odd lines were obtained from the image of the letter B. The resulting interlaced image is shown in the center. This composite image was then de-interlaced by separating even and odd lines, and filling in the empty spaces by copying neighboring lines of data
Fig. 2 Each video frame of the SLO angiogram contains the display of a video timer. This timer has a temporal resolution of 1/100 of a second
Non-interlaced data production
Methods Data source Fluorescein angiography data obtained from one normal, healthy patient receiving no systemic medications, who was participating in a prospective study already in progress and approved by an Institutional Review Board, was used for analysis. Informed consent was obtained and all tenets of the Declaration of Helsinki were followed. Video fluorescein angiograms were obtained using the Rodenstock SLO 101 (Munich, Germany) scanning laser ophthalmoscope. A 2.5 ml injection of sodium fluorescein 10% was given in a single bolus into a cubital vein followed by a saline flush. With the scanning laser focused on the ocular nerve head (ONH) and using a 40 ° image, the appearance of dye as it passed from arterial to venous circulation in the region of the ONH was recorded on super VHS videotape with a video cassette recorder (VCR). Throughout the test, the patient fixation was controlled by simple verbal instruction.
After data collection of both non-TBC and TBC video, a non-interlaced version of each was generated. After the digitized angiogram had been viewed, an appropriate section of video containing the arrival of dye in the artery and vein was selected. Using software developed in our laboratory, the interlaced frames were de-interlaced, or separated into their two individual fields. Interlacing is a technical video process in which the even lines of a video image are composed from a video image source, and then the odd lines are composed from the next frame of the source. (Fig. 1) Digitized video was then used to calculate arteriovenous passage (AVP) times of the fluorescein dye by previously described methods [2]. Statistical analysis The difference in AVP times before and after TBC on non-interlaced video was evaluated using a Wilcoxon signed-ranked test for comparison of nonparametric data. P values less than 0.05 were deemed significant. Also, the variation in elapsed time within a set of images was examined by calculating the elapsed time for 100 consecutive frames before and after TBC. The coefficient of variance was calculated and compared for statistical significance.
Video capture and time-based correction Video data was digitized with the Scion Image LG-3 frame grabber and Scion Image version 1.62 software (Frederick, Maryland) with a Macintosh G4 processor (Cupertino, California). Data was collected with and without serial S video connection of a Datavideo TBC-1000 (Taipei, Taiwan) single channel time-based corrector (TBC).
Results Time variation in consecutive frames The timer encoded on each video frame was used to determine the actual chronological order of the frames flowing from the video source (Fig. 2). Without TBC,
957
Fig. 3 The video timer reading from twenty five sequentially digitized frames are displayed graphically. Digitized without time-based correction, note that several of the digitized frames contained “old” images as suggested by timer readings jumping back and forth in time
Fig. 4 A graph of 25 consecutive frames in the video signal after the addition of time-based correction. Sequential order has been restored to the signal
digitized frames were not in correct chronological order (Fig. 3). After TBC, frames were collected in sequential order (Fig. 4). Before TBC, video was captured at a rate of 37.3 and 9.9 frames per second for non-interlaced and interlaced video respectively. TBC resulted in data captured at 31.2 and 15.6 frames per secondrespectively. Elapsed time variation within the same data set The elapsed time of a consecutive number of frames within a single TBC data set varied less than before TBC. Before TBC, the coefficient of variance was 7.57% and 51.3% for non-interlaced and interlaced data respectively. After TBC, the coefficient of variance was 0.15% and 1.69% respectively. Comparison of calculated and manual determination of AVP time The average AVP time error after TBC was shown to be significantly less (18.5%±31.2% vs 2.0%±1.2%;
p=0.0207). AVP time error is defined as the difference between calculated and manually determined AVP times.
Discussion The calculation of AVP times and the study of ocular hemodynamics with the SLO relies upon accurate and objective measurements. As always, there are limitations in the available technology that result in some reduction of accuracy, such as interlacing of the video images and the stretching and wear of the videotape. Our laboratory has extensively studied retinal and choroidal circulation using SLO angiograms and has encountered these problems in our analysis (Fig. 3). We employed two technical improvements which address these problems. One is the development of software which de-interlaces the video images, allowing for the separate analysis of each field and the removal of motion artifact. Another is the serial connection of a time-based corrector (TBC) between the VCR and the frame grabber hardware to eliminate timebased error.
958
Fig. 5 The first frame was digitized without time-base correction. There is a double image making analysis difficult. Digitizing with time-base correction stabilizes the frame. De-interlacing then separates the frame into its two respective fields (bottom images). Examination reveals that the second decimal place (black arrow) in the top frame is a composite of the bottom two digits
The addition of the video signal de-interlacing process was advantageous to the analysis of the SLO angiograms. One benefit is the elimination of the motion blurring that is commonly seen in rapidly moving objects. This is a problem with SLO films as we are analyzing very small retinal arteries and veins. It is very difficult for patients to keep their gaze totally fixed and even small movements result in motion blurring (Fig. 4). As described above, this phenomenon occurs because the two fields interlaced together to form a frame represent two different points in time [3, 10]. This is clearly seen in the examples of films. Before de-interlacing, careful examination of the timer on the top of the frame reveals that the hundredths position is a combination of two different numbers. After de-interlacing, the frame is separated into two images which differ in time by a few hundredths of a second (Fig. 5). Therefore, de-interlacing the SLO film doubles the number of frames and decreases the time lag between frames. This is demonstrated by the doubling of the acquisition rate from 15.6 frames/sec to 31.2 frames/sec after deinterlacing. Not only is motion blurring eliminated, but the image rate is also increased. This results in a more accurate assessment of the AVP time. Cardiac catheterization labs, which have encountered the same problems, have also been using non-interlaced video technology for some years with increased image quality and less motion artifact [3, 7, 8, 10, 14]. Another example of this technology being used is in the study of the motion of human sperm [6].
In order to understand exactly what a TBC is and what it does, a basic understanding of video recording is required. Each frame or field of video is composed of scan lines. For the image to be clear, each of these scan lines must begin at the same spot. However, video tape stretches easily when played and replayed around hot mechanical parts in a VCR. This results in a misalignment of scan lines, or time-based error [4, 9]. A TBC digitizes each scan line of video and stores it for a few microseconds and then releases it at a precise time, restoring the synchronization (sync) to the video signal [9]. This allows for the correction of time-based error and the display of a crisp image. It is also important to note that a TBC will not add time code or affect an existing time code system [4]. Therefore, serial connection of a TBC will not result in a loss of accuracy in the tape’s time code. Before the connection of a TBC, the frame grabber occasionally inserted frames out of sequence. This was only discovered after the addition of video signal deinterlacing to our data analysis increased the clarity of the timer in the image. Every few frames, the frame grabber would place a frame from its memory into the digitized signal rather than a new frame from the VCR (Figs. 1 and 5). Interestingly, it was often the same frame which began to be mistakenly inserted several times in a row before a new frame. After the connection of a TBC, this no longer occurred. It is possible that the Scion Images LG-3 frame grabber was overwhelmed by time-
959
base error and fed the contents of a previously filled frame buffer to the stored image sequence. As a TBC identifies and enhances the sync pulse [9], it enhances the signal and allows the frame grabber to digitize a stable video signal without time-based errors. This results in a smooth, accurate flow of images in the proper sequence. Time-based correction of the video also improved a number of quantitative parameters which we examined. The amount of time elapsed for a set of 100 frames was more consistent after TBC. The coefficient of variance was significantly reduced from 7.57% to just 0.15% (p<0.05). Additionally, the difference between calculated AVP times and manually determined times was significantly less after TBC. Both of these improvements may
be the result of increased stability in the video signal to the frame grabber. It is clear that the addition of TBC and de-interlacing increases the quality of the digitized SLO angiograms (Fig. 5). The increase in image quality speaks for itself. The increase in image quality and stability, as well as the increase in frame rate, enhances the accuracy of the determination of AVP times. The benefits far outweigh the expense of a TBC and the time associated with deinterlacing. Additionally, these techniques would be beneficial in many of the other applications of the SLO, such as choroidal filling time and capillary velocities. The improvement is dramatic. We now use both techniques for all of our analyses of SLO angiograms and recommend it to others using this machine for research purposes.
References 1. Duijm HFA (1997) Choroidal and retinal hemodynamics in glaucoma. Thesis Publisher, Amsterdam, pp 25–36 2. Harris A, Kagemann L, Cioffi G (1998) Assessment of human ocular hemodynamics. Surv Ophthalmol 42:509–33 3. Holmes D et al. (1986) Video X-Ray Progressive Scanning: new technique for decreasing x-ray exposure without decreasing image quality during cardiac catheterization. Mayo Clin Proc 61:321–326 4. Loehr M (1996) Edit suite: TBC trivia. Videomaker March: 22–26
5. Mainster MA, Timberlake GT, Webb RH, Hughes GW (1982) Scanning laser ophthalmoscopy. Clinical applications. Ophthalmology 89:852–857 6. Mortimer ST, Swan MA (1995) Kinematics of capacitating human spermatozoa analyzed at 60 Hz. Hum Reprod 10: 873–879 7. Seibert JA (1986) Improved fluoroscopic and cine-radiographic display with pulsed exposures and progressive TV scanning. Radiology 159:277–278 8. Seibert JA et al. (1984) Interlace versus progressive readout of television cameras for digital radiographic acquisitions. Med Physics 11:703–707 9. Sweetow S (1995) A TBC tale. Videomaker Sept: 61–67 10. Van Lysel M (2000) The AAPM/RSNA physics tutorial for residents; Fluoroscopy: optical coupling and the video system. Radiographics 20:1769–1786
11. Webb RH, Hughes GW, Delori FC (1987) Confocal scanning laser ophthalmoscope. App Opt 26:1492– 1499 12. Webb RH, Hughes GW, Pomerantzeff O (1980) Flying spot TV ophthalmoscope. App Opt 19:2991– 2997 13. Wolf S et al. (1989) Video fluorescein angiography: method and clinical application. Graef Arch Cli Exp Ophthalmol 227:145–151 14. Wondrow M et al. (1988) Technical consideration for a new X-ray progressive scanning system for cardiac catheterization. Cath Cardiovasc Diag 14:126–134