Disease_NN1 Ontology_NN1 
The_AT0 Disease_NN1 Ontology_NN1 (_PUL DO_NN1 )_PUR is_VBZ an_AT0 extensive_AJ0 ,_PUN hierarchically_AV0 organized_AJ0 vocabulary_NN1 of_PRF 12_CRD ,_PUN 694_CRD human_AJ0 diseases_NN2 that_CJT provides_VVZ a_AT0 framework_NN1 to_TO0 identify_VVI relationships_NN2 between_PRP diseases_NN2 and_CJC phenotypes_NN2 ,_PUN genotypes_NN2 ,_PUN and_CJC various_AJ0 other_AJ0 disease_NN1 attributes_NN2 ._SENT DO_VDB provides_VVZ semantically_AV0 consistent_AJ0 annotations_NN2 that_CJT allow_VVB one_PNI to_TO0 compare_VVI diagnostic_AJ0 evaluations_NN2 ,_PUN treatments_NN2 ,_PUN and_CJC patient-care_AJ0 data_NN0 over_PRP time_NN1 and_CJC between_PRP studies_NN2 ._SENT DO_VDB incorporates_VVZ concepts_NN2 and_CJC extensive_AJ0 cross-mapping_NN1 from_PRP standard_AJ0 clinical_AJ0 and_CJC medical_AJ0 terminologies_NN2 ,_PUN such_DT0 as_PRP Medical_AJ0 Subject_NN1 Headings_NN2 (_PUL MeSH_NN1 )_PUR ,_PUN International_AJ0 Classification_NN1 of_PRF Diseases_NN2 (_PUL ICD_NP0 )_PUR ,_PUN Online_AJ0 Mendelian_AJ0 Inheritance_NN1 in_PRP Man_NP0 (_PUL OMIM_NP0 )_PUR ,_PUN and_CJC the_AT0 National_AJ0 Cancer_NN1 Institute_NN1 Thesaurus_NN1 ._SENT The_AT0 ontology_NN1 provides_VVZ a_AT0 resource_NN1 to_TO0 connect_VVI genetic_AJ0 and_CJC phenotypic_AJ0 information_NN1 related_VVN to_PRP human_AJ0 disease_NN1 ._SENT 
Human_AJ0 Phenotype_NN1 Ontology_NN1 
The_AT0 Human_AJ0 Phenotype_NN1 Ontology_NN1 (_PUL HPO_NP0 )_PUR describes_VVZ phenotypic_AJ0 features_NN2 of_PRF hereditary_AJ0 ,_PUN congenital_AJ0 ,_PUN and_CJC acquired_AJ0 diseases_NN2 using_VVG a_AT0 structured_AJ0 and_CJC controlled_AJ0 set_NN1 of_PRF terms_NN2 ._SENT Although_CJS focused_VVN initially_AV0 on_PRP monogenic_AJ0 diseases_NN2 —_PUN about_PRP 50_CRD ,_PUN 000_CRD annotations_NN2 connect_VVB HPO_NP0 terms_NN2 to_PRP 4779_CRD diseases_NN2 in_PRP the_AT0 OMIM_AJ0 database_NN1 of_PRF genetic_AJ0 disorders_NN2 —_PUN HPO_NP0 now_AV0 includes_VVZ features_NN2 of_PRF more_AV0 than_CJS 3400_CRD common_AJ0 non-Mendelian_AJ0 disorders_NN2 ._SENT HPO_NP0 terms_NN2 can_VM0 have_VHI more_AV0 than_CJS one_CRD parent_NN1 in_PRP the_AT0 phenotypic_AJ0 hierarchy_NN1 :_PUN for_PRP example_NN1 ,_PUN podagra_NN1 (_PUL gout_NN1 of_PRF the_AT0 big_AJ0 toe_NN1 )_PUR has_VHZ parent_NN1 terms_NN2 gout_NN1 and_CJC abnormality_NN1 of_PRF the_AT0 foot_NN1 ._SENT HPO_NP0 has_VHZ been_VBN linked_VVN to_PRP OMIM_NN1 and_CJC to_PRP the_AT0 Orphanet_NP0 Rare_AJ0 Disease_NN1 Ontology_NN1 (_PUL ORDO_NN1 )_PUR to_TO0 increase_VVI the_AT0 interoperability_NN1 of_PRF phenotypic_AJ0 knowledge_NN1 in_PRP rare_AJ0 diseases_NN2 ._SENT One_PNI can_VM0 match_VVI clinical_AJ0 information_NN1 to_PRP phenotypes_NN2 at_PRP varying_AJ0 levels_NN2 of_PRF specificity_NN1 in_PRP the_AT0 ontology_NN1 's_POS hierarchy_NN1 to_TO0 formulate_VVI differential_AJ0 diagnoses_NN2 ;_PUN for_PRP example_NN1 ,_PUN a_AT0 clinical_AJ0 record_NN1 describing_VVG a_AT0 "_PUQ short_AJ0 2nd_ORD toe_NN1 "_PUQ would_VM0 be_VBI linked_VVN to_PRP the_AT0 HPO_NP0 term_NN1 short_AJ0 2nd_ORD toe_NN1 and_CJC its_DPS more_DT0 general_AJ0 terms_NN2 ,_PUN short_AJ0 toe_NN1 ,_PUN short_AJ0 digit_NN1 ,_PUN and_CJC abnormal_AJ0 digit_NN1 morphology_NN1 ._SENT 
Integration_NN1 of_PRF Ontologies_NN2 
Individual_AJ0 ontologies_NN2 can_VM0 serve_VVI as_PRP building_NN1 blocks_NN2 of_PRF broader_AJC ,_PUN more_AV0 general_AJ0 knowledge_NN1 resources_NN2 ._SENT By_PRP integrating_VVG related_AJ0 ontologies_NN2 ,_PUN knowledge_NN1 can_VM0 be_VBI shared_VVN and_CJC reused_VVN across_PRP domains_NN2 ._SENT The_AT0 US_NP0 National_AJ0 Library_NN1 of_PRF Medicine_NN1 's_POS Unified_AJ0 Medical_AJ0 Language_NN1 System_NN1 (_PUL UMLS_NP0 )_PUR Metathesaurus_NP0 seeks_VVZ to_TO0 provide_VVI semantic_AJ0 integration_NN1 of_PRF concepts_NN2 across_PRP ontologies_NN2 and_CJC vocabularies_NN2 ;_PUN a_AT0 single_AJ0 concept_NN1 unique_AJ0 identifier_NN1 (_PUL CUI_NN2 )_PUR in_PRP the_AT0 UMLS_NN2 Metathesaurus_NN1 may_VM0 refer_VVI to_PRP concepts_NN2 in_PRP several_DT0 component_NN1 vocabularies_NN2 ._SENT As_CJS described_VVN above_AV0 ,_PUN the_AT0 Foundational_AJ0 Model_NN1 of_PRF Anatomy_NN1 (_PUL FMA_NP0 )_PUR forms_VVZ much_DT0 of_PRF the_AT0 basis_NN1 of_PRF the_AT0 anatomical_AJ0 terms_NN2 in_PRP RadLex_NP0 ._SENT Knowledge_NN1 of_PRF radiological_AJ0 differential_AJ0 diagnosis_NN1 in_PRP RGO_NP0 has_VHZ been_VBN integrated_VVN with_PRP SNOMED_AJ0 CT_NN1 ,_PUN RadLex_NP0 ,_PUN DO_VDB ,_PUN HPO_NP0 ,_PUN and_CJC the_AT0 ORDO_NN1 ._SENT This_DT0 integration_NN1 allows_VVZ one_PNI to_TO0 pose_VVI new_AJ0 ,_PUN abstract_AJ0 questions_NN2 that_CJT relate_VVB diseases_NN2 and_CJC their_DPS imaging_AJ0 phenotypes_NN2 such_DT0 as_PRP ,_PUN "_PUQ Which_DTQ gastrointestinal_AJ0 system_NN1 diseases_NN2 may_VM0 cause_VVI an_AT0 abnormality_NN1 of_PRF the_AT0 genitourinary_AJ0 system_NN1 ?_SENT "_PUQ (_PUL Fig._NN1 4_CRD )_PUR ._SENT 
Limitations_NN2 
Although_CJS ontologies_NN2 provide_VVB a_AT0 powerful_AJ0 framework_NN1 to_TO0 organize_VVI information_NN1 ,_PUN they_PNP have_VHB several_DT0 limitations_NN2 both_AV0 in_PRP their_DPS design_NN1 and_CJC application_NN1 ._SENT With_PRP more_AV0 than_CJS 800_CRD ontologies_NN2 available_AJ0 through_PRP NCBO_NP0 BioPortal_NP0 (_PUL replaced-url_NP0 )_PUR ,_PUN it_PNP can_VM0 be_VBI challenging_AJ0 to_TO0 discover_VVI new_AJ0 or_CJC relevant_AJ0 ontologies_NN2 ._SENT Mappings_NN2 between_PRP these_DT0 ontologies_NN2 are_VBB sparse_AJ0 and_CJC few_DT0 are_VBB included_VVN in_PRP the_AT0 UMLS_NN2 Metathesaurus_NN1 ._SENT SNOMED_AJ0 CT_NN1 ,_PUN LOINC_NN1 ,_PUN and_CJC ICD-10-CM_NP0 are_VBB part_NN1 of_PRF the_AT0 Metathesaurus_NN1 ,_PUN but_CJC ,_PUN notably_AV0 ,_PUN RadLex_NP0 is_VBZ not_XX0 ._SENT Licensing_NN1 of_PRF ontologies_NN2 can_VM0 pose_VVI another_DT0 barrier_NN1 :_PUN the_AT0 US_NP0 federal_AJ0 government_NN1 licenses_VVZ SNOMED_AJ0 CT_NN1 for_PRP use_NN1 in_PRP the_AT0 USA_NP0 ,_PUN but_CJC the_AT0 ontology_NN1 is_VBZ not_XX0 openly_AV0 available_AJ0 worldwide_AV0 ._SENT Uneven_AJ0 or_CJC incomplete_AJ0 development_NN1 of_PRF ontologies_NN2 can_VM0 limit_VVI their_DPS use_NN1 :_PUN many_DT0 ontologies_NN2 ,_PUN including_PRP RadLex_NP0 ,_PUN are_VBB built_VVN either_AV0 through_PRP volunteer_NN1 effort_NN1 or_CJC are_VBB part_NN1 of_PRF a_AT0 larger_AJC initiative_NN1 ._SENT Only_AV0 a_AT0 minority_NN1 of_PRF RadLex_NN1 terms_NN2 include_VVB a_AT0 definition_NN1 ;_PUN new_AJ0 concepts_NN2 that_CJT enter_VVB the_AT0 vocabulary_NN1 of_PRF radiology_NN1 may_VM0 not_XX0 yet_AV0 be_VBI incorporated_VVN ,_PUN and_CJC there_EX0 are_VBB ongoing_AJ0 questions_NN2 of_PRF the_AT0 scope_NN1 and_CJC purpose_NN1 of_PRF any_DT0 ontology_NN1 ._SENT As_CJS ontologies_NN2 grow_VVB larger_AJC ,_PUN it_PNP is_VBZ increasingly_AV0 challenging_AJ0 to_TO0 assure_VVI that_CJT they_PNP are_VBB maintained_VVN appropriately_AV0 and_CJC that_CJT the_AT0 ontology_NN1 's_POS axioms_NN2 are_VBB logically_AV0 consistent_AJ0 ._SENT 
Few_DT0 ontologies_NN2 other_AJ0 than_CJS SNOMED_AJ0 CT_NN1 and_CJC LOINC_NN1 have_VHB been_VBN incorporated_VVN broadly_AV0 into_PRP commercial_AJ0 products_NN2 ,_PUN and_CJC solutions_NN2 adopted_VVN in_PRP industry_NN1 often_AV0 resort_VVB to_PRP simpler_AJC ,_PUN non-standards-based_AJ0 textual_AJ0 information_NN1 ._SENT Towbin_NP0 et_NP0 al._NP0 have_VHB highlighted_VVN the_AT0 widespread_AJ0 and_CJC substantial_AJ0 limitations_NN2 of_PRF traditional_AJ0 free-text_AJ0 body_NN1 part_NN1 labeling_NN1 in_PRP DICOM_NP0 and_CJC the_AT0 importance_NN1 of_PRF an_AT0 ontology_NN1 for_PRP body_NN1 part_NN1 labeling_VVG to_TO0 enable_VVI enterprise_NN1 imaging_NN1 ._SENT Smith_NP0 et_NP0 al._NN1 review_NN1 many_DT0 of_PRF the_AT0 challenges_NN2 of_PRF ontologies_NN2 for_PRP biomedical_AJ0 imaging_NN1 ,_PUN and_CJC propose_VVB a_AT0 coordinated_AJ0 suite_NN1 of_PRF ontologies_NN2 to_TO0 address_VVI the_AT0 needs_NN2 of_PRF biomedical_AJ0 imaging_NN1 ._SENT There_EX0 may_VM0 be_VBI lack_NN1 of_PRF incentive_NN1 in_PRP the_AT0 commercial_AJ0 space_NN1 as_PRP organizing_VVG or_CJC labeling_VVG data_NN0 ontologically_AV0 requires_VVZ more_DT0 work_NN1 and_CJC maintenance_NN1 ;_PUN it_PNP is_VBZ unclear_AJ0 what_DTQ might_VM0 motivate_VVI such_DT0 adoption_NN1 ._SENT 
Applications_NN2 
Natural_AJ0 Language_NN1 Processing_NN1 
Natural_AJ0 language_NN1 processing_NN1 (_PUL NLP_NP0 )_PUR has_VHZ been_VBN applied_VVN widely_AV0 in_PRP radiology_NN1 ,_PUN and_CJC can_VM0 perform_VVI more_AV0 efficiently_AV0 than_CJS human_AJ0 experts_NN2 with_PRP near-human_NN1 accuracy_NN1 ._SENT Many_DT0 clinical_AJ0 and_CJC research_NN1 applications_NN2 rely_VVB on_PRP automatic_AJ0 recognition_NN1 of_PRF medical_AJ0 concepts_NN2 in_PRP unstructured_AJ0 text_NN1 ,_PUN and_CJC the_AT0 accuracy_NN1 of_PRF recognition_NN1 can_VM0 affect_VVI the_AT0 analysis_NN1 of_PRF electronic_AJ0 health_NN1 records_NN2 ._SENT The_AT0 mining_NN1 of_PRF medical_AJ0 concepts_NN2 is_VBZ complicated_AJ0 by_PRP the_AT0 broad_AJ0 use_NN1 of_PRF synonyms_NN2 and_CJC nonstandard_NN1 terms_NN2 in_PRP medical_AJ0 documents_NN2 ._SENT Ontologies_NN2 have_VHB been_VBN shown_VVN to_TO0 improve_VVI performance_NN1 of_PRF NLP_NP0 applications_NN2 ._SENT 
Ontologies_NN2 can_VM0 support_VVI radiology_NN1 NLP_NP0 applications_NN2 by_PRP providing_VVG a_AT0 comprehensive_AJ0 vocabulary_NN1 of_PRF a_AT0 domain_NN1 ,_PUN along_AVP with_PRP the_AT0 terms_NN2 '_POS synonyms_NN2 ,_PUN abbreviations_NN2 ,_PUN and_CJC relationships_NN2 ._SENT Ontology_NN1 's_POS synonyms_NN2 ,_PUN abbreviations_NN2 ,_PUN and_CJC related_AJ0 terms_NN2 can_VM0 help_VVI identify_VVI diagnoses_NN2 and_CJC other_AJ0 entities_NN2 in_PRP radiology_NN1 reports_NN2 ._SENT For_PRP example_NN1 ,_PUN RGO_NP0 incorporates_VVZ 2957_CRD synonyms_NN2 (_PUL e.g._AV0 ,_PUN hepatoma_NN1 and_CJC HCC_NP0 for_PRP hepatocellular_AJ0 carcinoma_NN1 )_PUR that_DT0 can_VM0 improve_VVI recall_NN1 of_PRF an_AT0 NLP_NP0 search_NN1 routine_NN1 and_CJC can_VM0 also_AV0 help_VVI coalesce_VVB the_AT0 findings_NN2 into_PRP a_AT0 single_AJ0 unique_AJ0 entity_NN1 ._SENT Links_NN2 between_PRP ontologies_NN2 can_VM0 offer_VVI additional_AJ0 synonyms_NN2 ;_PUN for_PRP example_NN1 ,_PUN the_AT0 RGO_NP0 term_NN1 LEOPARD_NN1 syndrome_NN1 has_VHZ DO_NN1 synonyms_NN2 :_PUN Capute-Rimoin-Konigsmark-Esterly-Richardson_NN1 syndrome_NN1 ,_PUN generalized_AJ0 lentiginosis_NN1 ,_PUN Gorlin_NP0 syndrome_NN1 II_CRD ,_PUN lentiginosis_NP0 profusa_NP0 syndrome_NN1 ,_PUN Moynahan_NP0 syndrome_NN1 ,_PUN multiple_AJ0 lentigines_NN2 syndrome_NN1 ,_PUN Noonan_NP0 syndrome_NN1 with_PRP multiple_AJ0 lentigines_NN2 ,_PUN and_CJC progressive_AJ0 cardiomyopathic_AJ0 lentiginosis_NN1 ._SENT Hierarchically_AV0 organized_AJ0 information_NN1 allows_VVZ one_PNI to_TO0 search_VVI intelligently_AV0 for_PRP classes_NN2 of_PRF findings_NN2 ;_PUN for_PRP example_NN1 ,_PUN a_AT0 search_NN1 for_PRP phakomatosis_NN1 encompasses_VVZ more_DT0 specific_AJ0 terms_NN2 such_DT0 as_PRP neurofibromatosis_NN1 ,_PUN NF1_CRD ,_PUN and_CJC tuberous_AJ0 sclerosis_NN1 ._SENT 
Text_NN1 de-identification_NN1 tools_NN2 often_AV0 use_VVB algorithms_NN2 that_CJT attempt_VVB to_PRP redact_AJ0 people_NN0 's_POS names_NN2 by_PRP identifying_VVG words_NN2 that_CJT are_VBB capitalized_VVN ._SENT Radiology_NN1 reports_NN2 ,_PUN however_AV0 ,_PUN commonly_AV0 contain_VVB eponyms_NN2 for_PRP devices_NN2 ,_PUN findings_NN2 ,_PUN and_CJC diseases_NN2 that_CJT include_VVB capitalized_AJ0 proper_AJ0 nouns_NN2 (_PUL e.g._AV0 ,_PUN Foley_NP0 catheter_NN1 ,_PUN Kerley_NP0 B_ZZ0 lines_NN2 ,_PUN Wilms_NP0 tumor_NN1 )_PUR ._SENT Thus_AV0 ,_PUN de-identification_NN1 tools_NN2 may_VM0 require_VVI domain-specific_AJ0 vocabulary_NN1 so_AV0 that_CJT these_DT0 terms_NN2 are_VBB not_XX0 inappropriately_AV0 redacted_AJ0 from_PRP the_AT0 record_NN1 or_CJC replaced_VVN by_PRP arbitrary_AJ0 names_NN2 as_CJS in_PRP de-identification_NN1 techniques_NN2 such_DT0 as_PRP "_PUQ hiding_VVG in_PRP plain_AJ0 sight_NN1 "_PUQ ._SENT More_AV0 than_CJS 800_CRD of_PRF these_DT0 imaging_AJ0 findings_NN2 (_PUL such_DT0 as_PRP Rigler_NP0 sign_NN1 )_PUR have_VHB been_VBN incorporated_VVN into_PRP RadLex_NP0 ._SENT Connections_NN2 between_PRP ontologies_NN2 can_VM0 identify_VVI additional_AJ0 similarly_AV0 capitalized_AJ0 proper_AJ0 noun_NN1 synonyms_NN2 as_CJS alluded_VVN to_PRP in_PRP more_DT0 detail_NN1 above_AV0 ._SENT Although_CJS modern_AJ0 neural_AJ0 network_NN1 –_PUN based_VVN approaches_NN2 have_VHB reduced_VVN the_AT0 need_NN1 for_PRP feature_NN1 engineering_NN1 ,_PUN exposure_NN1 to_PRP controlled_AJ0 vocabularies_NN2 can_VM0 still_AV0 supplement_VVI these_DT0 techniques_NN2 and_CJC limit_VVB the_AT0 need_NN1 for_PRP extremely_AV0 large_AJ0 sets_NN2 of_PRF training_NN1 data_NN0 ._SENT 
In_PRP addition_NN1 to_PRP search-and-replace_AJ0 tasks_NN2 and_CJC named-entity_NN1 recognition_NN1 ,_PUN ontologies_NN2 can_VM0 supplement_VVI deep_AJ0 neural_AJ0 network_NN1 –_PUN based_VVN techniques_NN2 by_PRP creating_VVG word_NN1 embeddings_NN2 based_VVN on_PRP ontology_NN1 terms_NN2 ._SENT An_AT0 embedding_NN1 technique_NN1 converts_VVZ words_NN2 into_PRP vectors_NN2 ,_PUN or_CJC numerical_AJ0 representations_NN2 ,_PUN based_VVN on_PRP neighboring_AJ0 and_CJC co-occurring_AJ0 words_NN2 in_PRP a_AT0 text_NN1 corpus_NN1 ._SENT SNOMED2Vec_NP0 ,_PUN a_AT0 word_NN1 embedding_NN1 model_NN1 that_CJT uses_VVZ SNOMED_AJ0 CT_NN1 to_TO0 augment_VVI machine-defined_AJ0 relationships_NN2 ,_PUN can_VM0 outperform_VVI models_NN2 that_CJT are_VBB not_XX0 based_VVN on_PRP ontologies_NN2 because_CJS it_PNP strengthens_VVZ complex_AJ0 medical_AJ0 relationships_NN2 such_DT0 as_PRP "_PUQ breast_NN1 +_NN1 adenocarcinoma_NN1 →_NP0 invasive_AJ0 ductal_AJ0 carcinoma_NN1 "_PUQ ._SENT 
The_AT0 relationships_NN2 in_PRP an_AT0 ontology_NN1 enable_VVI more_DT0 robust_AJ0 medical_AJ0 reasoning_NN1 ._SENT The_AT0 multiple_AJ0 hierarchical_AJ0 levels_NN2 of_PRF terms_NN2 in_PRP an_AT0 ontology_NN1 enable_VVI connections_NN2 that_CJT go_VVB beyond_PRP word-list_NN1 associations_NN2 (_PUL e.g._AV0 ,_PUN for_PRP a_AT0 synonym_NN1 or_CJC proper_AJ0 noun_NN1 detection_NN1 )_PUR to_TO0 make_VVI an_AT0 end_NN1 model_NN1 more_AV0 medically_AV0 relevant_AJ0 ._SENT Researchers_NN2 have_VHB explored_VVN semiautomated_AJ0 approaches_NN2 to_TO0 learn_VVI ontologies_NN2 from_PRP text_NN1 ._SENT NLP_NP0 performance_NN1 was_VBD improved_VVN on_PRP a_AT0 gene_NN1 /_PUN protein_NN1 synonym_NN1 detection_NN1 task_NN1 by_PRP adding_VVG formal_AJ0 ontological_AJ0 knowledge_NN1 without_PRP modifying_VVG the_AT0 word_NN1 embeddings_NN2 ;_PUN the_AT0 ontology_NN1 provided_VVN context_NN1 that_CJT effectively_AV0 induced_VVN term_NN1 variability_NN1 while_CJS reducing_VVG ambiguity_NN1 ._SENT Biomedical_AJ0 ontologies_NN2 evolve_VVB continually_AV0 to_TO0 become_VVI more_AV0 robust_AJ0 and_CJC to_TO0 accommodate_VVI new_AJ0 imaging_NN1 techniques_NN2 and_CJC findings_NN2 ._SENT Ontologies_NN2 can_VM0 augment_VVI NLP_NP0 ,_PUN and_CJC NLP_NP0 can_VM0 be_VBI used_VVN to_TO0 analyze_VVI free-text_AJ0 corpora_NN2 to_TO0 find_VVI gaps_NN2 and_CJC to_TO0 suggest_VVI terms_NN2 to_TO0 add_VVI to_PRP existing_AJ0 ontologies_NN2 ._SENT 
Image_NN1 –_PUN Based_VVN Machine_NN1 Learning_NN1 
Image_NN1 –_PUN based_VVN machine_NN1 learning_NN1 can_VM0 be_VBI applied_VVN to_PRP a_AT0 virtually_AV0 infinite_AJ0 space_NN1 of_PRF potential_AJ0 applications_NN2 across_PRP diseases_NN2 ,_PUN findings_NN2 ,_PUN and_CJC imaging_NN1 modalities_NN2 ._SENT Early_AJ0 successful_AJ0 image_NN1 –_PUN based_VVN machine_NN1 learning_NN1 applications_NN2 have_VHB focused_VVN narrowly_AV0 on_PRP particular_AJ0 findings_NN2 or_CJC diseases_NN2 ;_PUN a_AT0 key_AJ0 challenge_NN1 is_VBZ to_TO0 identify_VVI the_AT0 applications_NN2 that_CJT will_VM0 have_VHI the_AT0 greatest_AJS impact_NN1 on_PRP radiologists_NN2 '_POS performance_NN1 and_CJC patient_NN1 care_NN1 ._SENT Ontologies_NN2 can_VM0 help_VVI guide_NN1 investigators_NN2 towards_PRP fruitful_AJ0 avenues_NN2 for_PRP research_NN1 ._SENT For_PRP example_NN1 ,_PUN musculoskeletal_AJ0 radiology_NN1 alone_AV0 entails_VVZ 635_CRD differential-diagnosis_NN1 listings_NN2 in_PRP RGO_NP0 ,_PUN such_DT0 as_PRP abnormal_AJ0 odontoid_AJ0 process_NN1 and_CJC premature_AJ0 craniosynostosis_NN1 ,_PUN with_PRP their_DPS interlinked_AJ0 imaging_NN1 findings_NN2 and_CJC diseases_NN2 ._SENT The_AT0 differential-diagnosis_NN1 listing_NN1 for_PRP abnormal_AJ0 odontoid_AJ0 process_NN1 links_NN2 to_PRP 37_CRD possible_AJ0 causes_NN2 ,_PUN including_PRP ankylosing_AJ0 spondylitis_NN1 and_CJC systemic_AJ0 lupus_NN1 erythematosus_NN1 ._SENT Similarly_AV0 ,_PUN in_PRP thoracic_AJ0 imaging_NN1 ,_PUN RGO_NP0 includes_VVZ 286_CRD imaging_AJ0 findings_NN2 caused_VVN by_PRP 2958_CRD diseases_NN2 (_PUL Table_NN1 3_CRD )_PUR ._SENT Understanding_VVG these_DT0 linkages_NN2 can_VM0 guide_VVI researchers_NN2 to_TO0 focus_VVI on_PRP findings_NN2 that_CJT are_VBB intertwined_VVN with_PRP diseases_NN2 to_TO0 increase_VVI the_AT0 chance_NN1 of_PRF substantive_AJ0 clinical_AJ0 impact_NN1 ._SENT 
These_DT0 hierarchical_AJ0 relationships_NN2 and_CJC mappings_NN2 can_VM0 then_AV0 help_VVI translate_VVI findings_NN2 that_CJT are_VBB detected_VVN using_VVG image-based_AJ0 techniques_NN2 to_PRP diseases_NN2 or_CJC other_AJ0 entities_NN2 which_DTQ may_VM0 strengthen_VVI an_AT0 image_NN1 –_PUN based_VVN machine_NN1 learning_NN1 model_NN1 and_CJC help_NN1 with_PRP one_CRD key_AJ0 area_NN1 of_PRF machine_NN1 learning_NN1 ,_PUN explainable_AJ0 or_CJC interpretable_AJ0 AI_NN1 ._SENT Image-based_AJ0 models_NN2 may_VM0 use_VVI saliency_NN1 maps_NN2 (_PUL "_PUQ heat_NN1 maps_NN2 "_PUQ )_PUR or_CJC activation_NN1 maps_NN2 to_TO0 explain_VVI which_DTQ image_NN1 features_NN2 contributed_VVN most_AV0 strongly_AV0 to_PRP certain_AJ0 findings_NN2 in_PRP an_AT0 exam_NN1 ._SENT Ontologies_NN2 and_CJC their_DPS relationships_NN2 can_VM0 then_AV0 help_VVI synthesize_VVI these_DT0 findings_NN2 to_TO0 propose_VVI causal_AJ0 relationships_NN2 or_CJC subsequent_AJ0 diseases_NN2 and_CJC outcomes_NN2 and_CJC can_VM0 show_VVI the_AT0 mappings_NN2 that_CJT led_VVD to_PRP these_DT0 conclusions_NN2 ._SENT For_PRP example_NN1 ,_PUN an_AT0 image-based_AJ0 AI_NN1 model_NN1 may_VM0 detect_VVI cirrhosis_NN1 ;_PUN ontological_AJ0 knowledge_NN1 can_VM0 suggest_VVI what_DTQ might_VM0 have_VHI caused_VVN the_AT0 finding_NN1 (_PUL alcohol_NN1 abuse_NN1 ,_PUN viral_AJ0 hepatitis_NN1 ,_PUN etc._AV0 )_PUR and_CJC possible_AJ0 consequences_NN2 (_PUL portal_NN1 hypertension_NN1 ,_PUN hepatocellular_AJ0 carcinoma_NN1 ,_PUN etc._AV0 )_PUR ._SENT 
One_CRD of_PRF the_AT0 most_AV0 challenging_AJ0 and_CJC time-consuming_AJ0 parts_NN2 of_PRF developing_AJ0 image_NN1 –_PUN based_VVN machine_NN1 learning_NN1 tools_NN2 is_VBZ labeling_VVG data_NN0 ._SENT Early_AJ0 approaches_NN2 typically_AV0 required_VVN cumbersome_AJ0 manual_AJ0 labeling_NN1 by_PRP domain_NN1 experts_NN2 ,_PUN but_CJC automated_AJ0 or_CJC semiautomated_AJ0 techniques_NN2 that_CJT generate_VVB large_AJ0 corpora_NN2 of_PRF slightly_AV0 imperfect_AJ0 training_NN1 data_NN0 can_VM0 approach_VVI and_CJC nearly_AV0 match_VVB the_AT0 performance_NN1 from_PRP carefully_AV0 manually_AV0 curated_VVN labels_NN2 ,_PUN although_CJS weakly_AV0 labeled_VVN datasets_NN2 generally_AV0 require_VVB significantly_AV0 more_DT0 labels_NN2 to_TO0 approach_VVI the_AT0 same_DT0 level_NN1 of_PRF performance_NN1 ,_PUN Ontologies_NN2 can_VM0 support_VVI the_AT0 development_NN1 of_PRF datasets_NN2 to_TO0 train_VVI or_CJC test_VVI image-based_AJ0 AI_NN1 models_NN2 by_PRP identifying_VVG mentions_NN2 of_PRF conditions_NN2 or_CJC imaging_NN1 findings_NN2 in_PRP radiology_NN1 reports_NN2 ._SENT 
High-quality_AJ0 NLP_NP0 models_NN2 can_VM0 facilitate_VVI important_AJ0 image_NN1 –_PUN based_VVN machine_NN1 learning_NN1 applications_NN2 and_CJC their_DPS evaluation_NN1 ._SENT Marrying_VVG an_AT0 image-based_AJ0 model_NN1 that_CJT detects_VVZ certain_AJ0 findings_NN2 or_CJC entities_NN2 with_PRP an_AT0 NLP_NP0 model_NN1 that_CJT can_VM0 accurately_AV0 identify_VVI those_DT0 findings_NN2 in_PRP the_AT0 report_NN1 could_VM0 be_VBI used_VVN to_TO0 automate_VVI peer_NN1 review_NN1 at_PRP a_AT0 scale_NN1 far_AV0 greater_AJC than_CJS what_DTQ is_VBZ feasible_AJ0 with_PRP current_AJ0 manual_AJ0 methods_NN2 ._SENT Machine_NN1 learning_NN1 tools_NN2 used_VVN by_PRP a_AT0 practice_NN1 should_VM0 be_VBI continually_AV0 monitored_VVN for_PRP performance_NN1 and_CJC for_PRP safety_NN1 events_NN2 ;_PUN parsing_VVG report_NN1 data_NN0 for_PRP the_AT0 findings_NN2 suggested_VVN by_PRP the_AT0 image-based_AJ0 model_NN1 can_VM0 help_VVI quantify_VVI the_AT0 performance_NN1 and_CJC potential_AJ0 errors_NN2 of_PRF both_DT0 AI_NN1 systems_NN2 and_CJC humans_NN2 ._SENT 
Radiomics_NN1 
Radiomics_NN2 can_VM0 link_VVI the_AT0 characteristics_NN2 of_PRF radiological_AJ0 images_NN2 with_PRP a_AT0 patient_NN1 's_POS genotype_NN1 ,_PUN gene_NN1 expression_NN1 profile_NN1 ,_PUN and_CJC prognosis_NN1 ._SENT The_AT0 use_NN1 of_PRF ontologies_NN2 can_VM0 help_VVI identify_VVI the_AT0 scope_NN1 of_PRF diseases_NN2 or_CJC imaging_NN1 findings_NN2 related_VVN to_PRP a_AT0 particular_AJ0 radiology_NN1 exam_NN1 and_CJC guide_VVB the_AT0 integration_NN1 of_PRF radiomics_NN1 features_NN2 with_PRP specific_AJ0 diseases_NN2 ._SENT The_AT0 Image_NN1 Biomarker_NN1 Standardization_NN1 Initiative_NN1 (_PUL IBSI_NP0 )_PUR has_VHZ defined_VVN uniform_AJ0 ,_PUN reproducible_AJ0 approaches_NN2 to_TO0 compute_VVI 174_CRD quantitative_AJ0 features_NN2 from_PRP medical_AJ0 images_NN2 ._SENT The_AT0 IBSI_NP0 radiomics_NN1 features_NN2 have_VHB been_VBN integrated_VVN with_PRP segmentation_NN1 algorithms_NN2 and_CJC imaging_NN1 filters_NN2 into_PRP the_AT0 Radiomics_NN1 Ontology_NN1 ,_PUN which_DTQ is_VBZ published_VVN through_PRP NCBO_NP0 BioPortal_NP0 ._SENT An_AT0 ontology-guided_AJ0 radiomics_NN1 analysis_NN1 workflow_NN1 has_VHZ been_VBN developed_VVN to_TO0 capture_VVI features_NN2 from_PRP imaging_AJ0 data_NN0 to_TO0 facilitate_VVI research_NN1 and_CJC clinical_AJ0 translation_NN1 of_PRF radiomics_NN1 ._SENT As_CJS radiomics_NN1 methods_NN2 become_VVB more_AV0 commonplace_AJ0 and_CJC sophisticated_AJ0 ,_PUN one_PNI can_VM0 imagine_VVI that_CJT particular_AJ0 computational_AJ0 features_NN2 will_VM0 become_VVI associated_VVN with_PRP various_AJ0 conditions_NN2 ,_PUN and_CJC such_DT0 information_NN1 could_VM0 be_VBI incorporated_VVN into_PRP a_AT0 diagnostic_AJ0 ontology_NN1 ._SENT 
Radiation_NN1 Therapy_NN1 Planning_NN1 
Ontologies_NN2 can_VM0 aid_VVI in_PRP planning_NN1 ,_PUN which_DTQ is_VBZ the_AT0 task_NN1 of_PRF identifying_VVG optimal_AJ0 sequences_NN2 of_PRF actions_NN2 to_TO0 achieve_VVI a_AT0 specified_AJ0 goal_NN1 ._SENT An_AT0 example_NN1 of_PRF planning_NN1 is_VBZ found_VVN in_PRP mapping_NN1 applications_NN2 that_CJT can_VM0 identify_VVI the_AT0 fastest_AJS route_NN1 to_PRP a_AT0 given_AJ0 destination_NN1 ._SENT In_PRP diagnostic_AJ0 radiology_NN1 ,_PUN a_AT0 planning_NN1 task_NN1 might_VM0 be_VBI to_TO0 identify_VVI the_AT0 most_AV0 cost-effective_AJ0 imaging_NN1 procedures_NN2 to_TO0 answer_VVI a_AT0 clinical_AJ0 question_NN1 ._SENT A_AT0 common_AJ0 planning_NN1 task_NN1 in_PRP radiation_NN1 oncology_NN1 is_VBZ to_TO0 select_VVI optimal_AJ0 radiation_NN1 ports_NN2 to_TO0 treat_VVI target_NN1 lesions_NN2 while_CJS sparing_VVG nontarget_AJ0 organs_NN2 ._SENT 
Two_CRD ontologies_NN2 have_VHB been_VBN developed_VVN in_PRP radiation_NN1 oncology_NN1 ._SENT The_AT0 Radiation_NN1 Oncology_NN1 Ontology_NN1 (_PUL ROO_NP0 )_PUR represents_VVZ clinical_AJ0 data_NN0 in_PRP the_AT0 radiation_NN1 oncology_NN1 domain_NN1 ;_PUN it_PNP incorporates_VVZ 1183_CRD classes_NN2 and_CJC 211_CRD properties_NN2 between_PRP classes_NN2 ._SENT The_AT0 authors_NN2 used_VVD the_AT0 ontology_NN1 to_TO0 query_VVI information_NN1 from_PRP various_AJ0 data_NN0 structures_NN2 without_PRP a_AT0 priori_NN2 knowledge_NN1 of_PRF the_AT0 schemas_NN2 of_PRF the_AT0 underlying_AJ0 relational_AJ0 databases_NN2 ._SENT The_AT0 Radiation_NN1 Oncology_NN1 Structures_NN2 (_PUL ROS_NP0 )_PUR ontology_NN1 comprises_VVZ 417_CRD anatomic_AJ0 and_CJC treatment_NN1 planning_NN1 classes_NN2 that_CJT describe_VVB commonly_AV0 contoured_AJ0 structures_NN2 for_PRP radiation_NN1 treatment_NN1 planning_NN1 ._SENT 
