bening? Refer for surgery. Indeterminate? (G). Is Patient a good candidate for early surgery? (H). 50% decrease in diameter after 3-6 months suppression? (F).
Graphical Style Sheets: Towards Reusable Representations of Biomedical Graphics Ramon M. Felciano & Russ Altman Section on Medical Informatics M.S.O.B. Suite x215, 251 Campus Drive Stanford University, Stanford CA 94305 +1 650 723-6979 {felciano, altman}@smi.stanford.edu ABSTRACT
We propose that the design characteristics shared by a family of data graphics can be represented as declarative, knowledge-based graphical style sheets that a generalpurpose visualization system can use to generate domainspecific data graphics automatically. Graphical style sheets (GSS) define the layout and drawing conventions shared by members of a particular family of data graphics. A GSS is a declarative mapping between Postscript-like graphical objects and biomedical data stored in object-oriented data structures. We describe the conceptual framework underlying our approach, and a prototype constraint-based visualization system (PALLADIO) and design representation language (P-SPEAK) we are building to evaluate this framework. Keywords
Visualization, Graphic Design, Health-Care Applications, Intelligent Interfaces, Constraint-based Systems, User Interface Toolkits, UIMS, World wide web, Electronic Publishing, Database Access INTRODUCTION
Over time, practitioners in a particular domain of biomedicine may custom-tailor generic statistical data graphics to display certain types of biomedical information. As these graphics become increasingly specialized, their value to biomedical practitioners increases, while their applicability to data from other fields diminishes. If these specialized graphics become widely used and accepted by practitioners in the field, they become as part of the lingua franca of that field. These domain graphics differ from generic visual formalisms [6] in their use of domainspecific layout rules, information symbols, and other drawing conventions in the display of domain data. Figure 2 shows several biomedical domain graphics. The World Wide Web (WWW) increasingly afford on-line publication and dissemination of biomedical data, including graphical presentations of this data. Because few user interface management systems support the creation of domain-specific interactive graphics, user interface designers must either forgo using these graphics or program them by hand. For example, SStructView is a Java user interface component that replaces conventional keyboard data entry in WWW-based interfaces to biomedical information systems with an interactive secondary structure
graphic (Figure 2C)[1]. We seek to facilitate the development of such interface components. METHODS
We propose to model domain graphic design conventions as declarative, knowledge-based graphical style sheets. Our conceptual framework includes (1) a Postscript-like drawing model based on points and drawing instructions applied to these points, (2) a frame-based representation of domain knowledge, (3) numeric and logical metrics that measure graphical characteristics (e.g., position, color), (4) constraints that represent target values for these metrics, and (5) graphical styles that map classes of data objects to sets of constrained metrics. A graphical style is a declarative representation of a graphic design preference that allows users to express what their visual design goals are without specifying how to achieve them. For example, a graphical style for secondary structure graphics (C in Figure 2) is ÒKeep all nucleotides that are part of a basepair 1 cm apart and draw a line between them.Ó We are building an implementation of this framework that includes PALLADIO, a system for visualizing data according to graphical style sheets, and P-SPEAK, a language for describing graphical style sheets. (Figure 1). PALLADIO uses general purpose layout algorithms and
Idea for Visual Presentation
Visual Design Preferences
Domain Knowledge (Concepts)
Style Sheet Editor
Domain Dataset
Generated Domain Graphic
Graphical Style Sheet (P-SPEAK) PALLADIO
Designer
G Reader
Evaluate
Figure 1 PALLADIO and Graphical Style Sheets. A Designer analyzes a domain graphic of interest, identifies salient visual elements and conventions used to encode data points, and describes these characteristics as design preferences in a GSS (written in P-SPEAK). A generalpurpose visualization system (PALLADIO) takes a domain dataset and the GSS as input, and generates a graphic that visually encodes data according to conventions in the GSS.
Patient with cold nodule on thyroid scan
+ 1
2
Do Aspiration
Bening?
Yes
50% decrease in diamter after 6 months suppression OR less than 1 cm? (F)
Begin suppressive Rx
Yes
+
No Repeat aspiration bening?
No
+ 2
1
Yes
3
No
Indeterminate? (G)
Is Patient a good candidate for early surgery? (H)
Yes
Yes
Refer for surgery
No
No
50% decrease in diameter after 3-6 months suppression? (F)
1
Yes
2
3
No Malignant?
Yes
Refer for surgery
Suppressive Rx and regular follow-up
No Inadequate Aspiration
+
Reevaluate (G)
(A)
+
+ 1
2
A A U G U U G C G C G C U GC CA A G U A CGA GC G U G UGCUCG A A U C G C U A G G G U C U G U C U U G U G C G G C G AU G A CG U A A GA GGA A A G U A A G G A U G G U U C C U U U AA U A U C AA G A C G U A U C A G U A UC C U A AG G
(B)
A
AU C U A U U CCUUG UCUUUAG U CC CA A A A GG GU AGGA A U A GA A A U C G U G G G U C U G A G A C A A U G G G G G C U A C G C G A G G A C U G C G UC A A A UA CU A A U G C C C C UU A GU U C
(C)
Figure 2 Biomedical Domain Graphics. (A) a clinical guideline flowchart outlining basic treatment steps for treating a patient with a cold thyroid nodule. Decision points are represented as hexagons, treatment steps as boxes, and patient state indicators as shadowed, rounded rectangles. ÒNoÓ decisions flow top-to-bottom, ÒYesÓ decisions flow left-to-right. (B) a pedigree tree used by epidemiologists to track inherited diseases: each symbol represents a family member, aligned vertically by generation, with solid symbols showing affected cases and open symbols showing unaffected cases. (C) a secondary-structure graphic showing partial structural information from a sequence of ribosomal (genetic) material (Paramecium tetraulia from [2], used with permission). Letters indicate individual nucleotides; Watson-Crick paired nucleotides are shown a fixed distance apart and with a line between them (or other symbols for non-canonical base-pairs); helices are shown as strands of parallel, base-paired nucleotides. Professional organizations may publish recommended standards for domain graphic designs; (A) is based on proposed standard from the Society for Medical Decision Making[9]. constraint solvers to generate a ÒgoodÓ graphic. PALLADIO creates a custom scoring function for each graphic based on the GSS and the elements in the dataset, and uses this function with available constraint solvers to produce an optimal graphic. We have used PALLADIO to lay out data using a GSS describing the design characteristics of the graphic in Figure 2C. We will perform three experiments to evaluate the contribution of this work: (1) we will evaluate the functional utility of graphical style sheets by evaluating the effectiveness of graphics generated by PALLADIO, (2) we will evaluate the expressiveness of P-SPEAK by building graphical style sheets for a range of domain-graphic families, and (3) we will evaluate the design flexibility of P-SPEAK by modeling real-world changes to a domain graphic over time as represented by a retrospective analysis of the historical evolution of the RNAseP secondary structure graphic. DISCUSSION
Our work has been influenced by previous efforts in visualization and constraint-based systems including APT[4], Juno-2[3], SAGE[7], and Multi-Garnet [8], as well as work on using fitness functions to generate good graph layouts[5]. Graphics generated by PALLADIO retain links to underlying data points. Thus, for any graphical object it is possible to determine what data point(s) influenced its creation and position on the page. We believe the automated generation of specialized biomedical graphics from declarative representations of design preferences will facilitate the use of such graphics as visual metaphors in biomedical user interfaces.
REFERENCES
1. Felciano, R.M., R.O. Chen, and R.B. Altman, R N A Secondary Structure as a Reusable Interface to Biological Information Resources. Gene-COMBIS, 1997. 190: p. GC59-GC70. 2. Gutell, R.R., Collection of small subunit (16S- and 16Slike) ribosomal RNA structures. Nucleic Acids Research, 1994. 22(17). 3. Heydon, A. and G. Nelson, The Juno-2 ConstraintBased Drawing Editor, 1994, SRC-131a, Digital Equipment Corporation: Palo Alto, California. 4. Mackinlay, J., Automatic Design of Graphical Presentations, in Computer Science. 1986, Stanford University: Stanford. 5. Marks, J., Automating the Design of Network Diagrams, in Computer Science. 1991, Harvard University: Harvard. 6. Nardi, B.A., A small matter of programming: perspectives on end user computing. 1993, Cambridge, MA: The MIT Press. 7. Roth, S.F., et al. Interactive graphic design using automatic presentation knowledge. in Human Factors in Computing Systems. 1994. Boston, MA. 8. Sannella, M. and A. Borning, Multi-Garnet: Integrating Multi-Way Constraints with Garnet , 1992, 92-07-01, University of Washinton: Seattle, WA. 9. SMDM Committee on Standardization of Clinical Algorithms, Proposal for Clinical Algorithm Standards. Medical Decision Making, 1992. 12(2): p. 149-154.
