This is one of the possible Use Cases.
The process of (semi)automatically identifying and labeling different parts of the brain cortex in neuroimaging applications, can be enhanced by the use of rule based applications. Such a process involves a great number of uncertain and vague information that need to be handled.
Originally proposed by: GiorgosStamou for the First F2F WG meeting as Use Case IVML. This work has been done in collaboration with Christine Golbreich and Ammar Mechouch (University of Rennes and INRIA-INSERM/Visages).
- Interest by many communities both commercial and academic like, medical (neuroimaging, diagnosis), multimedia processing, situation awareness.
Is any implementation effort underway? The IVML is currently implementing both fuzzy OWL reasoners and fuzzy Rule reasoners. Additionally, Fuzzy RuleML, http://image.ntua.gr/FuzzyRuleML/, also investigates reasoning with fuzzy rules.
3. Links to Related Use Cases
Situation Assessment and Adaptation: Requirement 5. RIF should include representation for uncertainty
4. Relationship to OWL/RDF Compatibility
OWL functionality is needed in order to express ontological knowledge about anatomical structures.
5. Examples of Rule Platforms Supporting this Use Case
- To the best of our knowledge no commercial rule platform supports this use case.
6. Benefits of Interchange
- Ability to use existing rule bases for the anatomy of brain cortex.
7. Requirements on the RIF
- Requirement 1. The RIF Core language should provide well-defined extensions that are intended in representing uncertain and imprecise knowledge, such as degrees of truth (partial truth) of propositions.
- Requirement 2. The RIF should have proper extensions for exchanging rules that represent uncertain knowledge, like rules with different confidences and/or atoms with different importance in the rule.
- Requirement 3. The RIF should have compatibility with OWL.
8.1. Actors and their Goals
- End-User: Wants to segment and identify the various parts of a brain cortex in an MRI (Magnetic Resonance Image).
- Medical-Expert: wants to encode knowledge about human anatomy in knowledge bases.
- Intermediate-Expert: wants to use existing knowledge bases to construct a domain specific knowledge base.
8.2. Main Sequence
- Step 1: Various Knowledge bases for human anatomy are created in the form of knowledge bases.
- Step 2: Knowledge from disparate knowledge systems is gathered and merged in a single rule base.
- Step 3: The rule base is used to (semi) automatically identify and label various parts of the brain cortex.
9.1. Fuzzy Reasoning with Brain Anatomical Structures
Medical image processing and analysis is a highly emerging research area in bio-informatics technology. Let us here consider the applications of decision support in neurology and neurosurgery by processing of MRI (Magnetic Resonance Imaging) images, which has gained considerable attention. The process usually involves two steps. In the first step the MRI image is automatically segmented into areas, each one associated with a set of features like the length and depth of a sulcus segment in a brain cortex, the connection of two sulcus segments etc. The second step involves the identification and labeling of the different parts of the brain cortex, based on the segmented parts. Between these two faces, the identification step is the most difficult one. Such a process can be assisted by knowledge-based tools, which provide both a rule component which captures the procedural part as well as dependencies between entities, and an ontology component that describes the entities of the brain anatomy. For example, using OWL the ontology part could contain the following entries,
OPIFGyrus subClassOF (some isdapartof IFGyrus)
IFGyrus subClassOF (some isfapartof frontallobe)
where OPIFGyrus represents the Orbital Pars of Interior Frontal Gyrus, IFGyrus the Inferior Frontal Gyrus and isdapartof represents the relation, isDirectAnatomicalPartOf. Furthermore, using OWL Lite one can capture the fact that, isdapartof is a sub-relation of a more broader relation, isapartof, that the relation hasdapart is an inverse of isdapartof and that isapartof is a transitive relation. On the other hand the rule component could be used to capture complex dependencies between the relations of the brain cortex structures. For example there are rules of the form,
separatesMAE(s,m_1,m_2) :- separatesMAE(s,m_1,sm_2), hasAnatomicalPart(m_2,sm_2), hasNoCommonParts(m_1,m_2), SF(s), MAE(m_1), MAE(m_2), MAE(sm_2),
where MAE represents a Material Anatomical Entity and SF a Sulcal Fold. Moreover, in many situations the existence of a part might be more important relative to others in the process of identifying the composite object, thus one might want to specify weights in the above atoms.
Now suppose that an image segmentation algorithm is applied to an MRI image in order to identify different brain parts. Since such algorithms cannot be sure about the membership or non-membership of an object to a certain concept, they usually provide confidence (truth) degrees. For example an algorithm might provide us with the information that, an object o_1 in the image isdapartof some other object o_2 to a degree of 0.8, o_2 isdapartof o_3 to a degree of 0.9 and that o_1 is an OPIFGyrus to a degree of 0.75. Using classical reasoning techniques, adapted to the fuzzy case, the membership degree of object to several classes and relations can be inferred.
A Technical Group on Fuzziness in RuleML has been established, http://image.ntua.gr/FuzzyRuleML/