5.1 Visualisation
Clinical diagnosis relies on the human visual capacity to
recognise structure and patterns. Effective visualisation is
becoming a critical aspect of turning biomedical data into
information and is becoming pervasive, particularly for
complex decision support applications (eg. Udupa and
Herman 1991).
Although commercial instrumentation to measure corneal
topography has become increasingly sophisticated,
manufacturers have been slow to provide effective
visualisation tools. As previously noted the most common
visual record, the photokeratoscope image, has limited use.
Line drawing representations of the topography such as relief
contours (Bonnet and Cochet 1962), wire mesh diagrams
(Itoi and Maruyam 1978), departures from sphericity (Clark
1974) and, more recently, colour coded contour maps of
corneal surface power (Maguire et al 1987, Klyce and
Wilson 1989a, Missotten 1994) have improved the clinical
value of the information but fall short of ideal because they
still do not effectively convey shape information for clinical
interpretation (Klyce and Wilson 1989b). More effective
presentation schemes are required.
For example, stereographic monitors could be used to
visualise the topographic data in three dimensions. The
scope of 3D visualisation in medical applications is
illustrated by the notion of virtual simulation (eg Rosenman
1991), in which the clinical design is carried out completely
on a virtual model. Although technical development and
implementation of 3D visualisation in medical
instrumentation has tended to lead clinical validation of the
real usefulness of the technology (eg Udupa and Herman
1991, Hsu et al 1993), its application is rapidly increasing
and the indications are that it can significantly improve
diagnosis and treatment.
Klyce and Wilson (1989a) used stereo-imagery to illustrate
corneal asphericity with stereo wire mesh diagrams. Not
surprisingly, they report a less than favourable response by
clinicians. The wire mesh models are unsatisfactory three
dimensional images and the complexity of mentally
interpreting a three dimensional departure from sphericity
renders them of little clinical value. Appropriate methods of
visualisation need to be developed.
5.2 Surface Matching and Difference Detection
A requirement of the final product is that selected corneal
models can be compared in order to:
i. study temporal changes in the corneal topography of an
individual patient,
ii. compare preoperative
topography, and
iii. compare corneal topography of an individual patient
with population models or an ideal corneal curvature
suited to a particular eye.
The cornea is however an uncooperative surface. The
corneal model does not contain any control points and there
are no natural targets. Any visible marks such as features on
the iris are difficult to identify and geometrically unreliable
because they are imaged through a refractive medium. There
are no reliable geometric entities such as a corneal apex or
corneal edges. Further, the patient's cornea cannot be located
and postoperative corneal
448
in repeatable positions or with accurately repeatable optical
fixation. Successive corneal models will not represent
exactly the same portion of the corneal surface. Any one
may contain only a subset of the others.
Among others, Rosenholm and Torlegárd (1988) have
investigated DEM matching without control points for
absolute orientation of =— stereomodels in aerial
photogrammetry and their methods have been applied in
close-range medical photogrammetry by Karras and Petsa
(1993). Methods of obtaining a least squares best fit surface
match without control have also been investigated by Pilgrim
(1988, 1991a, 1991b, 1992) and by Mitchell (1994, 1995),
Most methods minimise the difference in separation between
surfaces using a least squares solution. Some minimise the
angles between surface normals (eg Vezien and Koivunen
1993). In this application, where the clinician may be
primarily concerned with corneal curvature and corneal
power, it is possible that a matching technique that is
sensitive to surface normals will be more appropriate than
one which minimises surface separation. Reliable methods
of surface matching would significantly improve the utility of
any corneal mapping system.
5.3 Derived Quantitative Measures
Corneal curvature is an important derivative of the surface
model. Two valuable parameters are a global best fit radius
of curvature over the optic cap and a local estimate of corneal
curvature at selected points based on data from a limited
region. These data are used to calculate corneal power and to
characterise different human corneas. The number of points
that are required to model the cornea, the accuracy required,
and the trade offs between those two parameters have not
been determined elsewhere. Questions such as the following
must be addressed in order to design a model:
i. How many data points are required in order to compute
reliably the radius of curvature of the optic cap along a
defined meridian?
ii How many data points are required to compute reliably
the local radius of curvature of the cornea over any
small defined region?
iii. In each case, how accurate must the three dimensional
data be and what is the trade-off between accuracy and
sampling density?
The accuracy with which commercially available instruments
measure corneal power is invariably tested using spherical
targets. This is unreliable, since any instrument that uses
reflected mire techniques will be able to measure the radius
of curvature of a spherical target far more accurately than it
could a non-symmetric aspherical cornea. The accuracy
specifications for a digital implementation of the Keratocon
should follow from a proper analysis of the algorithms used
to compute parameters such as corneal power and a
consideration of the accuracy demanded by current
ophthalmic surgery. These issues are now being addressed
by some researchers. The indications are that an accuracy in
the order of -5um may be necessary for surgical procedures
such as photorefractive keratectomy (Missotten 1994).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
An acci
surface
ophthal
method:
accurate
when th
A photo
has bee:
reflectar
approxi
metric c
The in
measure
commer
grafted :
Full aut
problem
Keratoc:
the tar;
segment
experim
most lik
The opl
clinician
corneas
Given tI
instrume
videoker
impleme
only ad
accuracy
The key
incorpor
apply th
and visu
provide «
Keratocc
corneal
corneal |
Medical
practitio;
alone.
limited
temporal
disease,
measures
function:
power.
