You are on page 1of 4

10th Annual Florida AI Research Symposium, Daytona Beach, Florida, May 1997, 97-100

A LOGIC-BASED APPROACH TO PROSTHESIS DESIGN


*Peter Hammond and John C Davenport *Brunel University, UK and University of Birmingham, UK p.hammond@brunel.ac.uk;
Abstract The application of techniques from logic databases and definite clause grammars is illustrated in RaPiD , an intelligent design system used in prosthetic dentistry. design. However, there is a trend, as in other design domains, to encourage a more formal and a better documented approach to the design of medical devices. This has partly been encouraged by regulations introduced by the European Commission and disseminated by the UK Medical Devices Agency [3].

1 INTRODUCTION
This paper is concerned with the use of logic programming techniques in a knowledge-based system, RaPiD [1], for the design of a dental prosthesis called a removable partial denture (RPD). RaPiD uses ideas from logic databases, declarative graphics and critiquing, together with expert design knowledge, to provide a CAD-style graphical interface for both instructional and professional use. An RPD is a prosthesis for replacing missing teeth and related tissues. It restores the patients appearance, improves speech, assists mastication and maintains a healthy, stable relationship between the remaining natural teeth. The design of RPDs requires a detailed knowledge of design principles and clinical factors. It should therefore be carried out by a dentist after a careful analysis of the patient's oral condition and plaster models of the patients jaws. The resulting design is recorded as a diagram (annotated manually) and supplied to a dental technician to guide manufacture (figure 1a). In England and Wales, provision of RPDs increased from 228,000 in 1949 to 682,000 in 1994 at a cost in that year of $75M. Although the design of RPDs is a clinical responsibility, there is evidence from the USA, Australia, Scandinavia and the UK that it is frequently delegated by dentists to dental technicians who do not have appropriate clinical training. Poor RPD design results in failure of treatment and damage to oral tissues. An improvement in this unsatisfactory situation through the use of a knowledge-based system to make design more compliant with best practice could be cost effective and improve quality of health care. The design of RPDs is relatively informal when compared to general engineering or software

Figure 1a: A hand drawn RPD design


Stabilising element of clasp Reciprocating element of clasp Bounded denture base Circumferential clasp Rest acting as an indirect retainer for distal extension base Circumferential clasp

distal extension base

Bar clasp (Clasp Axis) Occlusal rest Major connector (lingual bar)

Figure 1b: A design produced with RaPiD

The expert design rules in RaPiD are based on a consensus established by survey of the 70 or so expert dental prosthetists in the 17 Dental Schools of the British Isles [2]. The graphical depiction of the design records and communicates important design decisions before manufacture and the expert guidance encourages best design practice. Thus, as well as being a useful design aid and educational tool, RaPiD also contributes to the standardisation of the design process.

Constraint representing design rule as a denial

2 A LOGIC DATABASE MODEL OF DESIGN


In RaPiD, an RPD design is depicted graphically (figure 1b) and is represented symbolically as a logic database [5]. The graphical and symbolic reasoning components are implemented in Prolog. The user's placement and manipulation of denture components are checked against constraints representing best design practice. A design alteration which contravenes any of these integrity constraints is critiqued so that the user is guided to a more acceptable design solution. Otherwise, the alteration produces an appropriate amendment to the graphical depiction of the design and the underlying logic database. Thus, a design is constructed dynamically by negotiation between system and user in such a way that its description and the static knowledge of component shapes and conformation are always consistent with the rules of expert design. We can capture this in the metalogical statement: Form User |- Design subject to the consistency of Form User Expertise where Form is the static knowledge of components and their assembly; Expertise is the static expert design knowledge; Design is the dynamic design negotiated between system & user; and User is the dynamic information supplied about a particular case. A particular design is represented as a set of clauses of predicate logic such as the following: tooth(t(1,2), artificial). tooth(t(1,4), present). saddle(saddle1, [t(1,3), t(1,2)] ). rest_on_tooth(rest1, t(1,4)). clasp_complex(clasp1, t(1,4), rest1). where t(Q,N) represents the tooth in position N (N {1,2,3,4,5,6,7,8} ) of quadrant Q (Q {1,2,3,4} ), according to a recognised convention. Expert design rules can be represented in a similar fashion. Below, we give a natural language version of a rule followed by its specification in predicate logic in the form of a denial. Design rule Supporting rests must be placed in a correct anterior/posterior relationship relative to abutment teeth

rest_on_tooth(Rest,Tooth) and centre_position(Rest, PointR) and supports_saddle_type(Tooth, Type) and correct_relative_position(Type, Position) and not relative_position(PointR, Saddle, Position). The user is informed by a critiquing message should any design update contravene the constraint.

3 LOGIC GRAMMARS AND SHAPE


For the automated generation of the shape of the major connector, the most important component and the most varied in shape, we employ a definite clause grammar. Tooth type natural artificial missing rest in 2nd pos rest in 1st pos periodontal problem onlay two rests
X

Icon

Token n a m n2 n1 x o n12

Table 1: icons and tokens in the grammar used for major connector shapes

This use of grammars in RaPiD was suggested during observation of the hand drawing of designs. For example, the juxtaposition of natural and artificial teeth and the presence of other components (rests, clasps and saddles) largely determine the shape of the border of the major connector as if they are laid out in a linear sequence, each producing a local constraining effect on its boundary. An arch of teeth is read clockwise and interpreted as a linear sequence taking into account the status of the teeth and the occurrence of rests. Table 1 contains icons and symbolic tokens used to represent elements in the edentulous arch. Thus, the lower arch depicted in figure 1b, when read in a clock-wise direction, corresponds to the sequence of icons

and to the list of tokens [m, a, a, n2, n, n, n, n, n, n, n, n, n2, a, a, n12]. The token sequence is parsed by the grammar which naturally enough is expressed in Prolog as a definite clause grammar. Some rules of the grammar are given in figure 2.

10th Annual Florida AI Research Symposium, Daytona Beach, Florida, May 1997, 97-100

arch --> free_end_start, nat_art_seq. arch --> free_end_start, nat_art_seq, free_end_finish. arch --> missing, nat_art_seq, free_end_finish arch --> missing, nat_art_seq, missing. missing --> []. missing --> [m], missing. free_end_start --> free_end_initiator, artseq. nat_subseq --> minor_connector_pair. nat_subseq --> natural. artseq --> [a]. artseq --> [a], artseq. minor_connector_pair --> [n2 , n1 ]. natural --> [n]. natural --> [n1 ]. natural --> [n2 ]. natural --> [n12]. natural --> [o]. natural --> [x].

corresponding approximately to the foci of an ellipse of a "similar" shape. The term inner(Point) describes the projection of Point into the inside area surrounded by the arch of teeth along a line bisecting the lines joining it to immediately adjacent foci.)

Natural2 X Artificial

Z Y

W V Natural1 U

inner(centre(last(Natural1))) inner(focus1(first(Artificial))) focus1(first(Artificial)) focus2(last(Artificial)) inner(focus2(last(Artificial)))

U V W X Y

inner(centre(first(Natural1)))

Figure 3: Graphical & symbolic form of seed points

Figure 2: Part of grammar used to generate shape of major connector

An augmented grammar (not shown) is used to produce a parse tree codifying the important features in the arch such as subsequences of natural or artificial teeth. This is used to generate a list of algebraic expressions describing the boundary points in terms of features of individual or groups of icons found. For example, the saddle (larger cross-hatched component) in figure 3 is identified by the parse subsequence natural(Natural1) saddle(Artificial) natural(Natural2) and generates the sequence U,...,Z of "seed points". The algebraic specification of these points is given in the bottom half of figure 3. (N.B. focus1(Tooth) and focus2(Tooth) should be interpreted as points within the boundary of Tooth

If these symbolic expressions are evaluated they generate the co-ordinates of points forming a crude outline for part of the connector boundary. Alternatively, left as they are, they can be manipulated symbolically to give more refined shapes or even be altered to comply with design rules governing the shape of the connector. For example, our survey of design experts confirmed support for the following design rule: Connector Rule A maxillary connector should cover the gingival margin when a single tooth separates a minor connector from a distal extension base. The rule is more easily appreciated in diagrammatic form in figure 4. It says that the space bounded by the lines connecting the points A,B,C,D,E should be absorbed into the connector shape. We can accommodate it by detecting the parse subtree components corresponding to the large shaded component and the two "rests" referenced by the points E and F in the figure below: free_end(Artificial) and minor_connector(Pair)

F A E

G D C B

F A

a flexible and accessible representation scheme for the shape of artefacts that can be specified in 2D. The use of grammar-based methods in RaPiD is quite specific to the application domain, at least in the current implementation. RaPiD has been in regular, routine use at the Birmingham Dental School for four years by both dental students and staff. An informal evaluation has been undertaken by prosthetic specialists, general dentists, dental students and dental technicians. We plan to test the transferability of the approach described here to other design domains in collaboration with industrial and clinical partners.

(a)

(b)

5 ACKNOWLEDGEMENTS
The RaPiD project was supported between 1992 and 1994 by the UK Higher Education Funding Council through their Knowledge-Based Systems Initiative. Between 1996 and 1999, RaPiD will be commercialised under the Teaching Company Scheme programme of the UK Engineering and Physical Sciences Research Council and the Department of Trade and Industry in conjunction with a software house TMS Ltd. References [1] Davenport JC, Hammond P, Potts AJC. A knowledge-based design system for removable partial dentures, Journal of Dental Research 70 (1991) 691. Davenport JC, Hammond P. The acquisition and validation of removable partial denture design knowledge - I: methodology and overview. Journal of Oral Rehabilitation. 23 (1996) 152-157; and II: design rules and expert reaction , 23 (1996) 811-824. EC Medical Devices Directive No 10. Guidance Notes for Manufacturers of Dental Appliances (Custom-made devices). Medical Devices Agency. Department of Health. London WC1B 5EP. Hammond P & Davenport J. Eliciting and modeling the design knowledge of multiple experts. Modelling Languages for KBS, Workshop at Vrije Universiteit, Amsterdam, January 1995. Hammond P, Davenport JC, Fitzpatrick FJ. Logic-based integrity constraints and the design of dental prostheses, Artificial Intelligence in Medicine. 5 (1993) 431-466.

Figure 4: The Connector Rule above says the shape in (b) defined by the path A, F, G is preferred to that in (a) identified by the path A,B,C,D,E,F,G.

which would ordinarily produce the sequence of points A,...,G shown graphically in figure 4a and symbolically in figure 5a. Instead we can generate a seed point sequence A, F, G that is more consistent with the design rule (see figures 4b and 5b). Manipulating the boundary seed points symbolically in this way until they conform to specified design rules also avoids unnecessary computation.

focus2(last(Artificial)) inner(focus2(last(Artificial))) inner(centre(first(Pair)))

A B C

[2]

inner(focus2(first(Pair)))
focus2(first(Pair)) focus1(second(Pair))

D
E F

inner(focus1(first(Pair)))
(a) corresponding to figure 4a

G
[3] A F G [4]

focus2(last(Artificial)) focus1(second(Pair)) inner(focus1(first(Pair)))


(b) corresponding to figure 4b

Figure 5: Seed points for boundary

4 Conclusions
RaPiD demonstrates some of the benefits of using a declarative approach to expert assisted design and to the representation of shape and conformation of components. The logic database model gives a clear and formal underpinning for a knowledge-based, critiquing style design tool and is transferrable to other domains. Logic grammars are

[5]

You might also like