KNOWLEDGE REPRESENTATIONS OF CONSTRAINTS FOR
PATIENT SPECIFIC IOL-DESTINATION
K. P. Scherer
Forschungszentrum Karlsruhe GmbH, Institut für Angewandte Informatik , P.O. Box 3640, 76021 Karlsruhe, Germany
Keywords: Knowledge based system, decision support, frame based representation, configuration system.
Abstract: A knowledge based system is a computer program, which simulates the problem solving process of a
human, who is an expert in this discipline. In the medical area there exist a lot of expert system components
for diagnosis of different diseases. In the ophthalmology, after a cataract surgical incision, the human lens is
removed and an artificial intraocular lens (IOL) is implanted. Because of many preconditions (patient
situation, operation technologies and IOL specifics) a knowledge based system is developed to support the
decision process of the IOL destination under the related (sometimes contradictory) constraints. A computer
aided IOL destination and decision support for a patient related individual optimized lens system can help to
enhance the life style of the bresbyope human. In comparison to classical software systems the heuristic
knowledge from the surgeons (aggregated over many years by research and clinical praxis) can be regarded
and user specific communication with the software system and an explanation part is available for the
decision making process. Useful representations of the situation are formalised knowledge representation
methods.
1 INTRODUCTION
Concerning reconstitute the vision quality for
presbyopes human eyes, a novel artificial
accommodation system (AAS) will be investigated
and developed in the institute of applied computer
science in the research centre Karlsruhe. This is
essential to guarantee life quality at an advanced age.
An optimized configuration and selection of such a
lens system can only guaranteed by a computer aided
decision support system. To do this, it is necessary to
generate and dispose of software structures for
storing and processing the information used in a
consistent manner (Scherer et al., 2006). Formal
representations have to be designed such, that the
information is accessed in a natural language. Based
on this acquired expert knowledge, new knowledge
can be produced and hypotheses and
recommendations can be validated or disproved. In
this proposal, the software components concern on
the selection and destination of parameterized
classical intraocular lenses. A transformation to the
ASS will be performed when the technical
development of the mechatronic system is nearly
finished.
2 BIOLOGICAL BACKGROUNDS
From the biological and medical point of view, a
mechatronic system is researched and developed for
the implantation into the capsule bag of a presbyope
Figure 1: Natural human eye as complex biological tissue.
human being after a cataract disease (Bergemann et
al, 2006). Fig. 1 shows a natural eye with a capsule
bag and an intraocular lens system before the
cataract operation. After extraction of the human
lens, the empty capsule bag should be filled with a
miniaturised complex self sustaining
290
P. Scherer K. (2008).
KNOWLEDGE REPRESENTATIONS OF CONSTRAINTS FOR PATIENT SPECIFIC IOL-DESTINATION.
In Proceedings of the Third International Conference on Software and Data Technologies - PL/DPS/KE, pages 290-294
DOI: 10.5220/0001869602900294
Copyright
c
SciTePress
accommodation system, which is nowhere existent
at time.
cornea
p
upil
anterior
chamber
lens
bulb
vitreous
lens shell
Figure 2: Abstract model of the human eye with the cornea
and the lens as refractive components.
Responsible for the refractive behaviour of the
human eye are the two components cornea with
different anterior and posterior surfaces and
furthermore the clear intraocular lens system (see
Fig. 2). The combination of these ray fractioning
surfaces leads to a more or less sharp image at the
retina. By replacing the human intraocular lens and
implanting a new technical lens system the
characteristic IOL features have to be configured in
such a way, that the desired refraction is performed
for each situation. This selection and component
characterisation must be performed by knowledge
based intelligent methods.
3 KNOWLEDGE DOMAINS
To destine a patient specific well parameterized
IOL, the following constraints must be regarded.
The eye parameter values, resulting from the
ophthalmologic measurement methods content
systematic measurement errors, which must be
corrected in an adequate manner. Furthermore the
IOL’s designed for implantations have fabrication
errors and must be accounted for the IOL parameter
calculation.
IOL
specifics
operational conditions
measure
ments
patient preconditions
relations
Figure 3: Different domain specific classes with relations.
Also any diseases before the cataract intervention
and specific surgeon dependent operational
techniques influence probably the results of the IOL
destination (Haigis et al, 2006). Additionally the
domains (patient preconditions, measurement
methods, fabrication related IOL features,
operational techniques) correlate among each other,
so the dependencies are very complex and they are
no longer linear (Fig. 3). A formalized description of
the different knowledge domains must be performed
(Fig. 4).
patient data:
- medical background
- physiological data
surgical conditions:
- operation techniques
Selection and characterization of an AAS, IOL:
knowledge representations
Figure 4: Knowledge based selection and destination of
the IOL implant.
4 FRAME REPRESENTATIONS
To describe the basic conditions for logical
reasoning a frame based approach is used. (Fig. 5)
Figure 5: Representation of a special class with attributes.
KNOWLEDGE REPRESENTATIONS OF CONSTRAINTS FOR PATIENT SPECIFIC IOL-DESTINATION
291
An extension to a so called concept is given by
evaluation of the possible attribute value by type
checking and range control (Fig. 5).
The class frame contains the following attributes:
Name/name: name of the class or of the object
Beschreibung/description: for more details and
specific features of the class
Fragetext/user question: relevant for the
knowledge acquisition concerning filling the
knowledge base
Fragetyp/type: specification of the question to
check the validity (real value or multiple-choice
or XML-text or other)
Wertebereich/range: the range of values for
checking correct input data
Autor/author: expert, responsible for the
knowledge
Quelle/source: source of the information
5 RULE BASED APPROACH FOR
IOL SELECTION
For the overall correlated information a rule based
approach is applied with the known Boolean
operators of the predicate calculus of first order. The
evaluation of the attributes in the classes patient
precondition, IOL characteristics, operational
conditions and the ophthalmic measurements form
the symptom tree, otherwise any information about
the selected IOL, the correlated IOL parameters and
IOL types compose the diagnostic tree.
symptoms
(
conditions
)
diagnostics
(
selection
)
IOL
characteristics
operational
conditions
logical inference
Selection of IOL
parameters
types
rules, functions, interpretations
measurements
patient condition
Figure 6: Correlation between symptoms and diagnostics.
6 DOMAIN SPECIFIC
KNOWLEDGE
REPRESENTATION
The modelling of the natural information about the
mentioned semantic topics means the development
of structures, which are formulised in class-subclass-
element relations. Along the relation “is a” between
classes and subclasses and elements an inheritance
mechanism is available. For the attribute slots,
constraints can be formulated. These frame based
approaches form a well defined hierarchical tree
structure for the new information system and are a
prerequisite for starting logical conclusions (Görz et
al, 2007).
6.1 IOL factors
One class-subclass tree concern the different IOL
specific factors, which are relevant for the selection
of a good implant. Not only the thickness of the lens
or design or fabrication related data are responsible
for the choose, also the used mathematical formulas
(Haigis, Hoffer, SRKII) leads to different IOL
parameters, because different preconditions in the
formulas are integrated (Fig. 7) (Findl et al, 2007).
IOL factors
Lens thickness
calculation
p
ro
g
ra
m
radians
desi
g
n
fabrication
Hai
g
is
Hoffer
SR
II
Figure 7: Class hierarchy of IOL factors (excerpt).
6.2 Diagnostics
For the implantation of the AAS or a classical lens
system different diagnostic methods are analysed. A
representation class is needed. The attributes of
these classes have both numerical and also linguistic
values. A basic a priori action of the ophthalmic
surgeons to destine a good lens system is the
ceratometry and the axis length measurement at the
patient’s eye. The measurement errors concern the
ICSOFT 2008 - International Conference on Software and Data Technologies
292
different submethods within the two necessary
measurements and they must be taken into account
for IOL destination (see Fig. 8).
Figure 8: Class hierarchy of diagnostics (excerpt).
6.3 Patient Situation
Also the patient specific situation and the history of
personal diseases and former refractive interventions
must be regarded for a correct destination of the
IOL. Patient specific diseases or an ametropia like
myopia or hyperopia and furthermore the age of the
patient or the presence of contact lenses affect the
possible measurement methods and so directly the
IOL destination (see Fig. 9).
Figure 9: Class hierarchy of patient specifics (excerpt).
6.4 Operational Conditions
Last not least the operational methods themselves
like the length of the intracorneal section or the
special location of the surgical intervention of the
eye gives a contribution to the postoperative
behaviour of the selected IOL. Therefore it must be
regarded before a selection is performed (Fig. 10).
Figure 10: Class hierarchy of operational features.
The knowledge entities are inputted into the class-
objects-attribute variables (see also frame based
structures). The knowledge acquisition can be
enhanced by consistency checking during the
acquisition step. Consistence means a validation of
values in a predefined numerical range or linguistic
values within a predefined term set or the number of
allowed values from the whole acquisition set.
This indirect constraint representation prevents
formally correct conclusions from being drawn on
the basis of data that are not allowed.
7 KNOWLEDGE ACQUISITION
Related to many national ophthalmic conferences in
Germany information and statements, published in
different literature are acquired in natural language.
The natural language based predicates are given to a
special medical expert for validation. The question
concerning each special medical statement could be
answered by the possible values
1) The statement is absolutely correct
2) The statement is absolutely wrong
3) Truth is given by a interval [0, 1] value as a
gradually evidence
4) The question is formulated in a wrong way
5) no comment to this knowledge statement.
The catalogue of questions concern the IOL
destination for a cataract patient, dependant on IOL
specifics, the operational conditions regarding
equipment failures and the patient specific situation.
The result of the interview and the comments of the
expert are following (the answers were given in this
first step of knowledge acquisition only by one
expert)
- 65 predicates were formalized.
- 20 statements were absolutely correct.
- 7 statements were absolutely wrong.
diagnostics
ceratometry
axis lenght measurement
manual
automated
ultrasonic Immersion optical
patient specifics
diseases
ametropia
glaucoma
strabism
myopia
hyperopia
ceratoplastics
a
g
e
depth prechamber
pupil width
contact lenses
operational influences
intrao
p
erativ
postoperative
section len
g
th
section location
one-
p
iece
multi
p
art
IOL
p
osition
KNOWLEDGE REPRESENTATIONS OF CONSTRAINTS FOR PATIENT SPECIFIC IOL-DESTINATION
293
- There was no gradually acceptance of information.
- 34 statements remain without any comment.
- 4 statements were formulated in a wrong way.
This procedure has to be performed again by other
experts and the set of formulas has to be enhanced.
But the analysis of the prototypical answers is very
interesting and the conclusion is following (here
only concerning IOL destination)
1) The knowledge concerning the IOL destination
including the different preconditions and
situations is distributed on different experts.
2) The knowledge is not available any time for
each human specialist.
3) The meaning can be contradictory.
The human expert expressions concerning the IOL
destination have to be transformed into rules.
Premises of the rules will be to the domain specific
class and object, conclusions will be results,
attentions and other IOL specifics.
8 EXPLANATION OF
REASONING
The warnings, recommendations and results must be
comprehensible. The explanation component is
responsible for providing the user with explanations
as to how the selection process was performed. It is
essential to understand the reasoning process and to
obtain new ideas for further advanced solution
processes. In this way, the confidence into the
conclusion and the developed knowledge based
system is enhanced.
Due to the probably very complex and wide
solution paths (numerous conditions and constraints)
the user specific results have to be analysed by
backward chaining processes. This requires a
comfortable capacity to interact with the system
through text and graphics.
9 CONCLUSIONS
The benefit and necessity of knowledge based
structures for selection of an optimized
parameterized accommodation system (ASS) are
outlined. The acquired knowledge (in this proposal
for the selection of a classical IOL) must be
managed using intelligent processes, because
absolute information is lacking in the instantaneous
state. Original rules developed may be rewritten
later and redefined. The complex knowledge is more
circular than linear. In this meaning comfortable
formulised structures and refinement mechanisms
must be developed as well as comfortable decision
making processes including their explanation
(configuration and selection).
Hence, the classical algorithmic based system has to
be extended to a knowledge based system for
selection and optimizing the IOL’s with following
advantages:
Natural language based user access
Object oriented knowledge structuring
Generation of network with causal relations
Consistent extension of knowledge base
Comfortable consistent refinement process
The advantage of a frame based information system
could be demonstrated. Formal representations are
designed for knowledge acquisition and the static
knowledge base, where an inference engine starts
with logical methods to select the desired patient
specific implant. Furthermore, causal relations and
interactive effects between the different
preoperational conditions have to be regarded in the
configuration and selection process.
REFERENCES
Scherer, K.P., Guth, H., Stiller, P., 2006, Computational
Biomechanics and Knowledge based Structuring of
Human Eye Components, International Congress on
Applied Modelling and Simulation, Rhodes, Greece,
June 26-28, 2006 ISBN 0-88986-561-2.
Bergemann, M., Gengenbach, U., Bretthauer, G., Guthoff,
R., 2006, Artificial Accommodation System – a new
approach to restore the accommodative ability of the
human eye. In: World Congress on Medical Physics
and Biomedical Engineering. Seoul.
Haigis W., IOL-Berechnung nach refraktiver
Laserchirurgie aus aktuellen Messwerten, in Klinische
Monatsblätter Augenheilkunde 2006, No. 223, Suppl.
1, 20. Jahrestagung DGII, 2006,
Puppe, F., Reinhardt, B., 2001, Multimediale wissens- und
fallbasierte Trainings- und Informationssysteme in der
Medizin, BMFT-Verbundprojekt, Projektträger DLR,
Förderkennz: 01EI9603/1,2001.
Görz G., Rollinger C.-R., Schneeberger J. (Ed.),
Handbuch der Künstlichen Intelligenz, in Fachbereich
1 „Künstliche Intelligenz“ der Gesellschaft für
Informatik e.V, Oldenburg Verlag München, Wien,
2007
Findl O., Biometry and intraocular lens power calculation
for cataract surgery, in Spektrum der Augenheilkunde,
Vol. 21, No. 2, Springer Verlag Wien, ISSN 09304282,
2007
ICSOFT 2008 - International Conference on Software and Data Technologies
294