Describing Localized Diseases in Medical Ontology: An FMA-based - PowerPoint PPT Presentation

Describing Localized Diseases in Medical Ontology: An FMA-based Algorithm Jean Charlet 1 , 2 , Laurent Mazuel 1 , 7 , Gunnar Declerck 2 , Patrick Miroux 5 , Pierre Gayet 6 1- AP-HP, Paris, France; 2- INSERM, U1142, LIMICS, 75006, Paris, France, Sorbonne Universités, UPMC Univ Paris 06, UMR_S 1142, LIMICS, 75006, Paris, France, Université Paris 13, Sorbonne Paris Cité, LIMICS, (UMR_S 1142), 93430, Bobigny, France; 3- Antidot, Lambesc, France; 4- Costech, Université de Technologie de Compiègne, Compiègne, France; 5- Dpt Urgences CHU Angers, France; 6- Centre hospitalier de Compiègne, France; jean.charlet@upmc.fr MIE 2014, Istanbul, September 1, 2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum An Information Retrieval Ontology 1 Lerudi Pathophysiology vs Anatomy Modelization 2 Problem Objective Material Algorithm 3 Result 4 Discussion 5 Perspectives 6 Jean Charlet et al. MIE2014

O NTOL U RGENCES : An information retrieval ontology In the context of the Le RUDI project Give physicians a GUI to facilitate the quick reading of the EHR for emergency care practitioners (mainly textual) Provide an overview of the patient in less than 2 mins Previous version (without FMA algorithm) available at http://purl.oclc.org/NET/spim/ontologies/public/ OntolUrgences/ A TOR, Terminological and Ontological Resource Identify concepts in the medical texts Thanks to terms denoting the concepts In order to reason about conceptual structures Thanks to “is-a” relation In order to display, among others, medical specialties in a concept cloud

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Le RUDI : Research screen Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum 'maladie abdominale' gastro-entérologie has subclass estUneMaladieDeLaSpecialiteMedicale(Subclass some) 'maladie du tube digestif' has subclass has subclass 'maladie du tube has subclass 'maladies de l'oesophage, de digestif global' l'estomac et du duodénum' 'maladie inflammatoire 'maladies de l'intestin grèle et localisee' du gros intestin' has subclass 'maladie du has subclass côlon' has subclass has subclass has subclass has subclass 'cancer intestinal' 'maladie du 'maladie du rectum' 'occlusion mésentère' digestive' has subclass 'maladies de l'appendice' chirurgie has subclass estUneMaladieDeLaSpecialiteMedicale(Subclass some) appendicite Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Le RUDI : Pathophysiology and anatomy (2/3) Pathophysiological and anatomical points of view play a key role in the characterization of medical pathologies The work of building a medical ontology must integrate and explicitly model them However, significant challenges had to be tackled in order to successfully achieve the modeling In addition, ensuring logical as well as semantic consistency when modeling both points of view added to this complexity Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Le RUDI : Pathophysiology and anatomy (3/3) Pathophysiology is the main axis of description of complex patients Pathophysiology is less well controlled than anatomy In most cases, pathophysiological position of a disease is unique while anatomical position is multiple (e.g. bronchopneumopathy) Develop the pathophysiological axis of the ontology and reuse a reference resource, the Foundational Model of Anatomy (FMA), to construct the localized diseases axis Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Objective Organize all the concepts of diseases in a tree 1 representing the pathophysiological perspective as understood by emergency medicine – e.g. CardiacAneurism is-a aneurism Develop Branch of localized diseases with defined 2 concepts (in order to position concepts defined in (1) through automatic classification). This requires: Define localization axioms – e.g. heart disease is on heart (and, if necessary only on heart) –, and Develop branch of anatomy necessary for the expression of the previous axioms – e.g. the heart itself: Heart is-a OrganWithCavitatedOrganParts . Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Used Resources Ontolurgences v3.03. 10,191 classes, 60 Object Properties, 1 Data properties, and 11,591 logical axioms including 11,339 subclass axioms, and 89 equivalent class axioms. OntolUrgences is built with core-ontology OntoMénélas 1 . The FMA. ( Foundational Model of Anatomy ) a reference ontology about human anatomy . 85,000 classes, 140 relationships connecting the classes, and more than 120,000 terms. We used the FMA v3.2.1 in OWL Full. 1 http://purl.oclc.org/NET/spim/ontologies/public/OntoMenelas/ Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Algorithm - 0 Annotation for Localization by medical practitioners Annotation of concepts of the ontology by an FMA identifier onto:pourFMA Possibility of AND (bronchopneumopathy is disease of the bronchi AND the lung) Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Algorithm - 1 Copying an extract of the FMA For each concept with an annotation onto:pourFMA (step 0), we copy concepts used in the annotation and all its relative superclasses to fma:Physical_anatomical_entity Built branch is stored under onto:StructureAnatomique (i.e. anatomical structure) which is a class of OntoMénélas core-ontology Conservation of FMA URIs Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Algorithm - 2 I Principle If 2 concepts are in meronymic relationships (part-whole), localized associated diseases are in subsumption relationships – e.g. “The mitral valve is a part of the heart” implies “the mitral valve disease is a heart disease” Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Algorithm - 2 II Creation of diseases hierarchy Given a concept X, if it has a relationship part_of to concept Y, then create the concept DiseaseOfY as father of the concept DiseaseOfX At each iteration, use one relation part_of , in order of preference systemic_part_of constitutional_part_of , then regional_part_of , To ensure the final connectivity of the graph, if the concept has no relation part_of , then follow the classical subsumption relationship Jean Charlet et al. MIE2014

Algorithm - 3 Localization definition by Defined Concepts EquivalentClass ( onto:DiseaseOfLung ObjectIntersectionOf ( onto:DiseaseOfLowerRespiratoryTract ObjectSomeValuesFrom ( onto:localizedOn fma:Lung ) ObjectAllValuesFrom ( onto:localizedOn fma:Lung )))

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Result Result: a new ontology 12,396 Classes, 60 Object Properties, 1 Data property and 17072 axioms, among them, 13332 subclass axioms and 3559 equivalent classes axioms Compare the last number to 89 equivalent classes axioms in the previous version Incrementality The algorithm is incremental. It is not necessary to keep a pre-algorithm version in order to apply it several times Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Discussion I Evaluation Evaluation of this work is done in-house and is used to improve the presented algorithm About FMA bugs During the development, this work uncovers FMA (small) mistakes. The FMA team corrects them Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Discussion II French vs English terms During the development of the algorithm, we work in English, because FMA is mainly in English. We collaborate with the FMA team and CISMeF in order to enrich the FMA with French terms (around 15,000) Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Perspectives I Which part_of ? The systemic point of view seemed the more understandable for the practitioner, because it corresponds to a partitioning of medicine into medical specialties. But the FMA team deleted systemic_part_of due to its lack of coherence. We need to make new tests using, in order of preference, constitutional_part_of then regional_part_of About complexity of created ontology The resulting ontology is very complex. We investigate the possibility of reducing the visibility of some concepts (or deleting them) in the ontology Jean Charlet et al. MIE2014

An I.R. Ontology Physiopath. vs loc. Algorithm Result Discussion Perspectives Addendum Perspectives II About onto:pourFMA Need to have annotations which point to “more” systemic anatomy elements Current work: re-localization of 2,000 concepts in order to correspond to medical specialties New annotation onto:pourFMAO in order to better take into account the localization on X OR Y Jean Charlet et al. MIE2014