A Test Phantom for Retinal Screening
Bahman Kasmai, Margaret Flatman
Norfolk & Norwich University Hospital NHS Trust

A few months ago, a colleague handed me an article by Dr. Peter Scanlon to read. The article was titled “ Digital Cameras and Related Software, The Problem we face”. Dr Scanlon highlighted a number of problems. When I read the article, first I was very impressed with his depth of knowledge and understanding of these very important issues, secondly it induced a sense of déjà vu in me . I felt that I had encountered these issues in the world of medical physics and radiological imaging where images are mostly produced by ionization radiation as against non-ionizing visual light.

Dr Scanlon has made some recommendations on resolvability and calibrations. Most of these recommendations have parallels in the world of radiological imaging except the one in the recommended useful resolutions, or sampling frequency as we prefer to call it. In the world of diagnostic imaging, while it might be practical to recommend a resolution, this is usually avoided because these figures tend to loose their relevance over time as the imaging technology evolves. This also applies equally to image compression technology used for storage and display of diagnostic images. Dr. Scanlon’s recommendation on the uses of a model eye as a “Golden Standard” is very sound. When implemented, it will remove the need for specifying the useful resolution and the image compression technology. I think the primary requirement to resolve objects of minimum 30 microns should be adequate for the purpose. In order to device a “golden standard” we need to go back to the basics and see what we mean by resolution and how the resolution is determined.

Resolution is defined as the degree of discernable detail. The unit of measurement for Resolution is line pairs per mm (lp/mm). A line pair is one dark and one light bar - one cycle. The methods used for measurement of resolution, is by visual inspection of standard targets such as the USAF 1951 resolution chart or Siemens Star Resolution Chart. For an objective measurement of an optical system performance, the minimum MTF values (MTF stands for Modulation Transfer Function) or MTF-curves are used. The MTF curves are provided by the optical system manufactures to give an objective measure of quality for optical resolvability. For visual light imaging systems, test objects are used extensively for calibration and quality assurance purposes. In the world of radiological imaging it is mandatory to check the spatial and contrast resolutions for imaging on a routine basis as part of a comprehensive QA programme. The QA in the world of radiology imaging is very advanced and mature. One service in the radiology world that is very close to the retinal screening service is the national mammography screening. The QA programme for mammography is the most developed in the radiological imaging. I understand that the NSC is planning to develop a quality assurance programme for retinal screening. I am sure many lessons could be learned from the Breast Screening QA model. If a model has not already been developed, I think the mammography screening QA model could provide a good template for QA in Retinal Screening.

QA in Radiology is mature and well developed in terms of protocols, organization and resources. Each region has a quality assurance director and a quality assurance reference centre. The national coordinating committees produce guidance on good practice and set standards and targets for staff working in the programme. Comprehensive QA Guidelines exists for each staff group such as screeners, technicians and physicists.

Screeners, technicians and physicists have many test object phantoms as tools for measurement and QA monitoring.

Using phantoms or test objects, technicians and physicists measure and monitor many parameters such as contrast, spatial resolutions, sensitivity and uniformity. Many of these parameters could equally apply to Retinal Screening. To do similar measurements and monitoring in retinal screening, a Phantom or Model eye will be needed. Is there a model there that one can build on?

As far as I can tell, the first model eye was designed by Ray Applegate, the wavefront-sensing pioneer. It was called Applegate Model Eye. It was designed as a “gold standard” for reporting aberration characteristics of eye.

What do the suppliers of the retinal screening cameras use for their own in-house testing? As I understand, Haag-Streit, the main supplier of the Canon cameras, use a proprietary test object with an image of retina at the back. When questioned, they were very guarding of their design details. All they could tell us was that they use an image of retina in the enclosure shown. Tocon, another major supplier uses a model eye manufactured by Heine, called Heine Retinoscope Trainer, as a test object. Heine is an optical system manufacture in Germany. I have been in touch with the Heine’s head of R&D and had a very positive response in my proposal to modify the Retinoscope Trainer for use as a test phantom for retinal screening.

The first time a model eye was used for measurement of spatial resolution was by Peter Koch Jensen of Denmark in 1999. He successfully used the Zeimer & Mori Model of eye, a model designed for teaching and practicing laser therapeutics, to measure the limiting resolution of fundus cameras against direct ophthalmology. One interesting feature of this model eye is that it allows replaceable targets, a desirable and necessary design feature for a future retinal screening phantom.

We have put together an experimental retinal phantom as a design demonstrator with our own very limited workshop facility. The phantom is designed to fit easily into the camera by an operator for QA monitoring. We have learned through experiments that the phantom design has to be flexible enough to adopt to default settings used for routine day to day screening.

Our experiments also showed that the conventional black and white test object used for lenses would not produce desired pictures for default flash intensity settings and can produce undesirable optical effects. It is my understanding that the fundus cameras are specifically optimized to provide images of retina with its own combinations of RGB colours. The retinal pathologies usually have very low contrast with surrounding tissues and therefore the design for a test object must reflect this fact.

We took a representative sample of retinal images and looked at their histograms to choose a suitable colour as typical background for a series of test images. We looked at the maximum pixel counts and colour values for each of the three primary colours and come up with a dominant colour for the background with RGB values of R=253, Blue= 63 and Yellow= 107.

This is an example of a test image for quantitative measurement of both spatial and contrast resolutions. To determine the contrast resolution one could count the number of discernable circles of graduating contrasts. We think this type of test images are useful tools for the QA of display monitors. We all know that display monitors can drift, lose lamps, get dirty, and sometimes are not set up properly to begin with. The ultimate goal is of course to have such test images as target s within a retinal screening phantom. The final design for a retinal phantom should allow replaceable targets.

A retinal screening phantom is a valuable tool for:
 Standardizing image quality for screening centres
 Evaluating image quality in a quantitative manner
 Ensuring optimum performance for day to day use
 Assessing different manufacture’s claim
We would like to propose that as part of a Quality Assurance programme, a design specification for a test phantom be developed and communicated to potential optical system manufactures for an ultimate production of a NSC approved test phantom.