Midterm

1. Biology of the heart (20 points)

During class we have used the lung and respiratory system as an example of the kind of information that is associated with classical biology. Part of the goal was to provide you with a framework for organizing information about other anatomical structures. Such a framework in anatomy class has been called a Knowledge Organization Template (KOT). Like a frame or object in programming, the attributes of such a KOT depend on the type of the anatomical structure. Part of the KOT for the class “organ” is as follows:

Name
Synonyms
Definition
External appearance (surfaces, shape, features)
Internal anatomy (parts)
Input/Output (blood, nerves, lymph, additional depending on the organ)
Relationships (location, adjacent organs, continuities)
Histological structure (kinds of tissues and cells)
Function

Your job is to fill in these attributes for the heart, in 1-2 pages, but no more. These pages can include up to two illustrations. You may use any source, including books from the library and online resources, but you should cite these sources at the end of the entire exam. The illustrations can be from any source, but the source of the illustrations must also be cited.

The values of the attributes should just be text, augmented by up to two illustrations. Note that the same structure can appear in more then one attribute value. In the case of function you should briefly describe the cardiac cycle, that is, the sequence of events that cause blood to flow into and out of the heart.

2. Querying the FMA and the semantic web (20 points)

In the remaining questions we will explore how to represent and use in computer applications the type of information described in question #1.

As discussed in class one type of anatomical information is symbolic information about concepts and relationships, of which part-of relationships are particularly difficult because there are so many ways to divide something up. In this exercise we will determine parts of the heart that are of a particular type, and will look at the tradeoffs between obtaining this answer using 1) a browser interface like the FME and 2) a semantic web query engine.

We will limit our query to parts of the Wall of the heart that are of type epithelium.

a. (5 points) Explain the difference between constitutional part and regional part in the FMA. Explain what is meant by transitive closure of one of these types of parts?

b. (5 points) Use the FME to look at the transitive closure of the constitutional parts of " Wall of heart". Which of these parts are of type "Portion of epithelium"? List the parts.

c. (5 points) Write a vSparQL query to answer this same question, and provide a screenshot of your query and your results. As in the previous class assignment you should modify the query on that page and paste your modified query into the vSparQL demo text box at http://fma.biostr.washington.edu:8080/FMA_OWL_Client/. Your query should be pasted below all the existing lines that start with "PREFIX" on that page. Hints: All names in the FME, such as "Temporal lobe" are represented in the vSparQL query like fma:Temporal_lobe, and the subclass relationship is specified by rdfs:subClassOf. Given these hints your main challenge is to figure out how to add an "and" condition to the WHERE clause in the sample queries. Looking at the paper by Detwiler should help.

d. (5 points) Discuss the tradeoffs between these two methods of arriving at the answer. Are the results the same in both cases. Why or why not? Why and when would you prefer one method over the other?. Given the current state of the query engine, are these tradeoffs the same whether the query is about a small part of the heart or the entire heart?

3. Imaging (20 points)

The AnnoteImage program is an example of a tool for labeling regions on images. As shown in this screenshot the user loads an image, manually draws contours around regions of interest, and associates a name with each region. The resulting annotations are saved in a separate XML file associated with the image.

a. (5 points) In the current downloadable version (version 1) the names are simply typed into the program. However, a Java web start demo shown on the AnnoteImage project page (version 2) queries the vSparQL ontology web service to retrieve a pick list of names. Why might you want the user to be able to select from such a controlled terminology when creating the labels? What are the tradeoffs between allowing the user to type arbitrary text and requiring them to pick from a list? Is there an advantage to generating the list via a query rather than simply picking from a list of all possible terms?

b. (5 points) The current demo (version 2) of AnnoteImage has a single hardcoded image, and a hardcoded vSparQL query. Describe the design of AnnoteImage version 3, in which intelligent use of queries generates different lists of structure names from which to pick depending on the body region of the image and its spatial extent (eg macroscopic versus microscopic). How would the user enter the information? Given this information describe in words the query to be generated. As an alternative, given that a single structure has been annotated, what other kinds of relationships in the FMA could be used to come up with the likely next structures to be annotated?

c. (5 points) Suppose you want to make a version 4, in which the user does not have to do all the outlining by hand, but wants to make use of image processing. Given the name of a structure, describe the steps that would be involved to automatically segment the structure in the given image. How would the user interface be modified to allow automatic segmentation? How would the user correct errors? What sort of knowledge would be useful in aiding the segmentation?

d. (5 points) Finally, consider the design of an image retrieval system based on the annotated images created by AnnoteImage. How could an ontology like the FMA be used to generate "intelligent" queries that retrieve images annotated with anatomical terms that are not explicitly stated in the query? Give an example (in words) of such an intelligent query. How could content-based retrieval be used in addition to the use of the controlled terminology?

4. Simulation (20 points)

Annotations with controlled terminology from an ontology can be useful, not only for searching across image databases, but also for searching across model databases as a first step towards model integration. Assume that computational models have been annotated using the methods described in the Gennari paper, and that the annotations for each model are saved as instances of an ApplModel. Sketch the design of a model database that would use the ApplModel annotations as a means for finding relevant models stored in the database. How would such a system answer queries like, "Retrieve all models that are related to the cardiovascular system", or "Retrieve all models that involve blood flow in the aorta?"

5. Google Earth for the Human Body (20 points)

One of the topics you will hear about throughout your informatics education is the notion of a personal health record or PHR. Recently, IBM research described a prototype 3-D body-based PHR application that maps individual patient health information to a 3-D body model. Such an application, which has been advocated by many groups including our own Structural Informatics Group, has been likened to a Google Earth for the human body. Yet there are many problems that make such an application much more challenging than Google Earth.

Compare and contrast Google Earth versus a potential Google Earth for the Human Body. What is the same about the two applications? What is different? Why is the human body much more difficult? What are some of the specific problems that would need to be solved, and how could some of the methods we have discussed to-date help to meet these challenges?

Finally, why would one want to build such an application? Would it be better than a text-based PHR?