ARVO / ISIE 2012 Image Quality and Disease Screening

Image Quality and Disease Screening Performance Comparison of
Clinic-Based and Telemedicine-Based Retinal Cameras
ARVO / ISIE
2012
MK Smolek,1,2 N Notaroberto,1-3 A Jaramillo,3 JG Diamond,4,5 OR Batlle5
1CLEVER
Eye Institute, Pearl River, LA; 2Louisiana Eye Research Institute (LERI), Pearl River, LA;
3EyeCare 20/20, Mandeville, LA; 4Southeast Louisiana Veterans Health Care System, New Orleans, LA;
5Tulane University Department of Ophthalmology, New Orleans, LA
Introduction
Cost, efficiency, image quality, and disease
sensitivity are important when choosing a retinal
camera for telemedicine. The Centervue DRS
(Centervue SpA, Padova, Italy) is a robotic
camera designed for telemedicine, but are the
low resolution DRS images adequate for retinal
disease screening? We compared the DRS to a
full-featured Canon CX-1 (Canon USA; Lake
Success, NY) for screening performance and
image quality.
Views of the 5.1 Mp DRS (left) and the 15.1 Mp
Canon CX-1 (right). Note the lack of operator
controls on the DRS, while the CX-1 uses manual
controls to align, focus, and acquire the image.
DRS
CX-1
DRS
CX-1
DRS
CX-1
DRS
CX-1
DRS
CX-1
DRS
CX-1
DRS
CX-1
DRS
CX-1
Methods
Full-color 40° to 45° central-field photographs of
25 eyes from 13 randomly chosen patients with
various types of ocular pathology were captured
using both cameras. One eye of one patient was
not used due to a technical fault not noticed at
the time of capture. Screeners were masked as
to which camera produced each image, as well
as to the patient history. The gold standard was a
retinologist (NFN) with access to the full patient
history and who made the original diagnosis.
PART 1: Three retinologists (AJ, JGD, & ORB)
screened for signs of pathology in each image
that was presented randomly and independently.
PART 2: The three retinologists and a vision
scientist (MKS) used a forced-choice image
quality grading method (better, worse, equal) to
compare the images from both cameras.
PART 3: ImageJ software (NIH) was used to
analyze the RGB pixel intensity histograms of
each image to assess the image quality in terms
of three contrast-related parameters.
All results were tabulated and analyzed using
SigmaStat (SPSS, Inc) or a web-based Kappa
calculator (http://justusrandolph.net/kappa/)..
Results
Discussion
Examples of some of the retinal images used in the study are shown
above. DRS images are shown on the left and CX-1 images on the right.
The CX-1 has three times the resolution of the DRS, however, none of the
clinicians indicated that this difference was relevant to their ability to screen
several different types of diseases (drusen, retinopathy, macular degeneration,
hypertension, retinal vein occulsion, histoplasmosis, etc.). Statistically, the
screening with the DRS was as reliable as screening with the CX-1 camera.
PART 1: The free-marginal Kappa statistic was 0.51 for both the DRS
and the CX-1, indicating identical screening performance for both
systems. The majority of screening differences were among those cases
with mild forms of pathology. The ability to perform an accurate screening
with either system was limited because the patient history and other
clinical data were not provided to the screeners.
PART 2: The subjective image quality of the DRS was scored as
better than the CX-1 by all 4 graders (Wilcoxon Signed Rank test; P
<0.016 to P <0.001). The most experienced retinologist chose the DRS
quality as superior for every image. The other graders indicated that
several image-pairs were equal, but the DRS was superior in the majority.
PART 3: The mean pixel intensity (MPI), standard deviation of MPI,
and the maximum pixel intensity from the ImageJ analysis were all
significantly higher with the DRS images (paired t-test; P <0.001).
Results are presented in the 3 graphs in the upper right. The objective
measurements of contrast and brightness were in agreement with the
subjective appearance of image contrast and brightness found in Part 2.
All of the clinicians noted that the DRS had better clarity and contrast than the
CX-1. All of the clinicians assumed the DRS images were from the full-featured
clinical camera. The DRS appears to have image algorithm enhancements and
filtering that enhance the contrast and cut through media opacities better than
the CX-1. In addition, the manual CX-1 focus caused image quality problems.
We did find that the DRS was occasionally affected by eyelid artifacts. We
believe the DRS could be improved as a telemedicine camera by an image
quality analyzer that would signal if an image needs to be retaken due to small
pupils or eyelid/eyelash artifacts.
Financial
Disclosure
MKS: Board of Directors, LERI
Research Director, CLEVER
NFN: Board of Directors, LERI
AJ :
None
JD:
None
OB:
None
Conclusions
Screening performance was equal for both cameras. However, images from the
DRS retinal camera were preferred by all 4 graders because of enhanced
contrast, brightness, and clarity, and because of the consistent focus, which
made the screening process easier and faster.
All rights reserved. Copyright © 2012 CLEVER Eye Institute.