Generation Congenital Heart Defects in Pediatric Cardiology Kartik Vishwanath, Venkatesh Viswanath, Yugyung Lee, Ph.D School of Computing and Engineering University of MissouriKansas City William Drake, MD Children's Mercy HospitalKansas City
2 OntoDiagram Application for pediatric cardiology that generates Mullins like diagram based on domain descriptions, clinical observations and measurements Mullins AtlasA collection of widely used diagrammatic representation of anatomy of heart structure (Mullins Diagram) Mullins Diagrams are useful in hospital reporting tools as they efficiently represent the defects Given a description of heart by domain experts, the system should automatically generate the Mullins diagram representing the conditions as described 3 Sample Mullins Diagrams Pulmonary Atresia Tricuspid Atresia Complex TOF TAPVR LSVC Why Generate Diagrams? 5 Situs and Relations: There is levocardia with visceral and atrial situs solitus, atrioventricular concordance (D-looped ventricles) and normally related great arteries {S,D,S} noteVenous Connections: There are normal systemic and pulmonary venous connections, with the superior and inferior vana cavae returning normally to the right atrium, and all four pulmonary identified returning normally to the left atrium. noteAtria: The left and right atria have normal chamber size, structure, and relations. noteAtrioventricular Valves: The mitral and tricuspid valves have normal structure, size, placement and function. There is no mitral insufficiency and only physiologic tricuspid insufficiency. noteVentricles: There is normal left ventricular chamber size and normal subjective left ventricular funtion with flattened interventricular septal motion. note There is normal right ventricular chamber size. note The right ventricle wall thickness is moderately increased. note There is normal global left ventricular function. note There is flattened interventricular septal motion. noteOutflow Tracts: The left ventricular outflow tract has normal size and geometry, without stenosis or narrowing. note The right ventricular outflow tract has moderate hypoplasia. note There is no subaortic conus. note There is marked right ventricular outflow tract turbulence (stenosis). note The peak right ventricular outflow tract velocity is 4.1 meters/second. note The peak right ventricular outflow tract gradient is 67 mmHg. noteSemilunar Valves: The aortic valve is normal, with three normal leaflets, normal mobility and no prolapse. note There is non-turbulent flow and a normal flow velocity across the aortic valve with no evidence of stenosis. note There is marked hypoplasia of the pulmonary valve annulus. note The pulmonary valve annulus measures 2.1 mm in the long axis outflow view. note The peak velocity across the aortic valve is 1.3 meters/second. note The pulmonary valve leaflets are moderately thickened. noteAortic Root Ascending Aorta: The aortic root, including the sinuses of valsalva, sinotubular ridge, and proximal ascending aorta are normal without stenosis, narrowing, or dilation. notePulmonary and Thoracic Arteries: There was supravalvular pulmonary stenosis present. note There was hypoplasia of the main pulmonary artery. note The left pulmonary artery appeared hypoplastic. note The right pulmonary artery appeared hypoplastic. note There was a right-sided aortic arch present. note There was a patent ductus arteriosus visualized. note There was a right- sided patent ductus arteriosus present. note There was left-to-right shunting demonstrated across the patent ductus arteriosus. note There was a right aortic arch with a retroesophageal segment and left descending aorta forming a vascular ring. noteSeptal Defects: There is a malalignment type ventricular septal defect present. note The ventricular septal defect measures 12 mm from the apical four-chamber view. note There is a moderate degree of aortic override. Fibro-annular continuity of the aortic and mitral valves is present. note The aorta has about 50 per cent override. note Comprehending textual descriptions is time consuming! 6 Diagram is easier to understand! Complex TOF 7 Motivation Visual representation is intuitively easier to understand than text information. Pediatric Cardiology faces new or variations of heart defects Mullins atlas are comprehensive but not complete Difficult to search and retrieve similar diagrams Domain experts manually draw diagrams by hand or modify existing ones in Mullin's atlas 8 Pulmonary Artery Aortic stenosis Tricuspid insufficiency Normal Bicuspid valve Mitral valve cleft Coarctation Right-left shunt PDA Patent Foramen Ovale Bifurcation of PA anterior Overriding Aorta True PA Parachute Mitral Valve hypoplasia above below Hidden behind Component sheared Scaled along long axis connections missing translation transposition translation Ontological Framework Diagram World Domain World Domain Experts Perspective Diagram Perspective Heterogeneity of Perspectives 9 Challenges Heart Structuresimple & complex Mapping domain knowledge onto diagram specifications Perspective and representational gap between medical domain and diagram domain Domain Modeling Modeling congenital defect information Modeling anatomical structure Implicit domain knowledge Anatomical consequence of defects Diagram Modeling Individual component modeling Spatial orientation between components Transformations on components to reflect changes 10 Heart Structure Simple & Complex Simple An anatomically simple structure Intuitively hierarchical part-of relationships (left heart, right heart, etc.) Heart components are less in count Anatomically new additions are very less and predictable Complex Variations of each component are large Variations of heart as a whole are diverse Complex relationship between various anomalous conditions
11 Simple Complex 12 OntoDiagram: Focus Translation from domain description to diagram
Domain Rules Diagram Ontology Mapping System 14 Walk through the System 15 Ontologies An ontology is an explicit formal specification of the terms in the domain and relations among them (Gruber 1993). To share common understanding of the domain knowledge on congenital heart defects among cardiologists, nurse, database engineer or software agents To make domain assumptions explicit so that defect associations patterns could be extracted To separate domain knowledge from operational knowledge To map different perspectives of domain and diagram 16 Domain Modeling Modularizing domain knowledge Congenital Heart Defect Ontology [UMLS] Classification of defects, anomalous conditions Definition for domain terms like stenosis, atresia, etc. Anatomical consequences of defects Association between defects Anatomy ontology [FMA] Multi-perspective classification of heart anatomy Possible anatomical changes in each heart component Orientation between components Measurement and Diagram Ontology [LOINC] 17 Diagram Modeling Diagram Models Component Association Model (Spatial Orientation between components) Structure Model (Component hierarchy) Conversion Model (Transformations) Annotation Model (Relevant annotations) Diagram perspective of changes Abnormal growth Missing parts Transposition Transformation Combination
18 Component Association Model Spatial relations with components and their neighboring components Six-tuple {Ct, Cn, D, A, O, P} where Ct and Cn components, Dspatial orientation of Ct with respect to Cn, Ainterface presence, Oorientation, Ppresence of component e.g. {Descending Aorta, Aortic Arch, bottom, attached, below, present}
bottom top right top top top right right 19 Structure Model The heart is considered to be made of two layers The first layer consisting of the chambers of the heart (wall of atria and ventricles) The second layer consists of the other components of heart (Pulmonary artery and Pulmonary valves, Aorta and Aorta valves, etc.) The relationships between the first and the second layer components 20 Interface Points & Gate Points ComponentComponent LayerLayer 21 Conversion Model Transformation of components Geometric transformation (e.g., Scale, Rotate, Translate) Polynomial transformation (e.g., Shear) Domain specific transformation (e.g, dilation, coarctation) Transformation model defined as {C T ,T, P} C T component being transformed Tname of the transformation Ptransformation parameters
22 Hierarchical Composition Hierarchical model closer to domain perspective
Faster diagram generationfew components change
Optimal selection of components 23 Hierarchical Composition Steps Step 1: Configuration 1. Identify an appropriate set of components and their abstraction level to be composed [Congenital Defects and Association models] 2. Determine the components to be transformed [Conversion Model] 3. Determine the relationships between components (below, above) [Component Structure Model] Step 2: Composition Starting from the most specific level to the root 1. Transform component images using the transformation operations if necessary (scale, rotation, etc) 2. Rank the components [Image Model - Color Schema] 3. Compose the components using interface points [Structure Model] 4. Forward the interface points to the upper level Step 3: Annotation 24 Diagram Composer Congenital Heart Defect Ontology Anatom y Ontolog y Measurement Ontology Diagram Ontology Mapping System Domain Description Diagram Instruction Files Domain ontologies Ontological Framework OntoDiagram System 25 OntoDiagram Prototype Constructed domain ontologies using the protg in OWL. Implemented the query interface in Java using the Jena. Implemented the image composition module using the Java Advanced Imaging package.
26 Sample Output (Patent Ductus Arteriosus) 27 Conclusion The domain descriptions of the congenital heart defects are mapped to generate a diagrammatic representation of defects. Ontology based domain modeling Component based diagram modeling Rule based mapping between domain and diagram Hierarchical composition of components
A prototype system is currently under testing at the Childrens Mercy Hospitals and Clinics facilitate existing treatments for managing patients with severe heart disorders.