MediGrid
– sharing credible knowledge and reliable resources
Kryštof Slabý, Petr Lesný
University Hospital Motol, Prague
MediGrid
Documenting and implementing biomedical computations in the network environment
Utilizing the GRID distributed networking concepts
Documentation goals Maximize reusability of the algorithms
SearchTrust
Allow interoperability of the algorithmsData identity
Provide basis for future publishing of new algorithmsPeer reviewed publicationConnection to established resources
Implementation goals
Usable both on local computer and in network environment
Distributed and reusable Possibility of automated workflow
construction Security, Accounting, Access control
MediGrid: current status
Developed general mechanisms of documentation
Documented algorithms from biomedical domains:Paediatric auxologyNephrologyPulmonary function testsDNA diagnostics
Contents Introduction
Documentation of medical algorithms
Trust
Phenomenological Description of Biomedical Data
Grid Networking Closing remarks
„Sharing knowledge in biomedicine“
Evidence Basedness
Workflow
Ad-hoc ontology
Next?
Sharing knowledge
What You share? Formalized description
How do You find it? Ontology
Do You trust author? Review No => References to literature
Formalized description
administrative (title, author, date published etc.)
„syntax“ (datatype, range, input, output)
„semantic“ (description, references, method of measurement, units ...)
technical (WS address, unittests)
Knowledge lifecycle
Publication
Formal criteria Review
Proper description Evidence based (references)
Responsibility author, reviewer, EBM
Formalized Computer processable, not understandable
Evidencebasedness
Knowledge expressed Literature based
How solid is the basis?
Relation of referenced to description Quality of description
Next?
Algorithmic medicine
ComputationsDose_of_drug = Dose_per_kg × Weight_kg
ScoresE.g. Glasgow coma scale
Guidelines If high glucose, then examine blood K+ level
Pros and Cons
GoodObjectivization helps young M.D.s to deal with
complicated situationsCan simulate „expert opinion“Can be automated (saves time)
BadMust be utilized in proper clinical situationGarbage in – garbage outCan „dehumanize“ medicine
Documentation
Human readableDescription, Argumentation, …Semantics („meaning of the data“)
Human and computer readableLinks to external entities, …
Computer readableSyntax of input and output dataConstruction of user interface…
Documentation
Algorithms and Input and output data
BMI = weight / height2
1. Body mass index
2. Body weight
3. Body height
4. The BMI calculation
Documenting algorithms
Name, author, abstract, definitions of terms List of inputs and outputs Method of computation or processing Supportive and Controversial expert opinions Restrictions References to peer-reviewed journals Classification systems – UMLS, MeSH
Documenting inputs and outputs
Name, author, abstract, definitions of terms
Precision, Interpretation Controversy Origin of data, Measurement References to peer-reviewed journals Classification systems – UMLS, MeSH
Documenting inputs and outputs
1. BSA = 4.688 * ((weight) ^ (0.8168 – (0.0154 * log10 (weight)))) [Boyd, 1935]
2. BSA = 3.207 * ((height)^(0.3)) * ((weight) ^ (0.7285 - (0.0188 * log10 (weight)))) [Milazzo, 1985]
3. Drug_dose = BSA * dose_per_sq_m [Drug data sheet]
Can be the result of (1) or (2) utilized in (3)?
Documentation service
Set of documentation – algorithms and their inputs and outputs
Centralized × Distributed
Next?
„Trust“ in science
Implicit concept of Scientific method in biosciences
Basic levels of trust Highest: What I have personally experimented with Very high: Information from textbooks ;-) High: Publications in peer-reviewed journals Middle: Reviews in peer-reviewed journals Lower: Publications in other scientific journals Low: Posters etc.
„Trust“ in science
ModificationsTrust / Distrust in Someone [‘s work]Delegation of trustRelation to my own experiments
Trust: Documentation
Documentation must be reliable1. Peer-review systém of experts
2. Connection to peer-reviewed journals
3. Possibility of trust delegation
Trust: Implementation
We work with patients‘ data
Network layer (e.g. SSL)Do we communicate with the right computer?
Next?
Modules
= relations among entities some computable
weight & height => BMI => obesity
Workflow
Encapsulates multiple modules Same interface as single module Static/dynamic construction Workflow editor
Example of complex workflow
BP < NIBP
MAP
MAP
RR detector
PEP < BPPEP
BPS/BPD
Inst. fH
ECG
MAP estimate
HR
HR
Next?
What is the problem?
BMI = weight / height2
Is it possible to document the „semantics“ of these data?Body mass indexBody weightBody height
Semantic web approach
Define complex „ontology“unit of measurementrelations to body functions and dimensions is_a, is_part_of, …
InstrumentsProtégé, OWL, PROLOG
ExamplesBody weight in kg is_a body weightBody weight in lb is_a body weight
Semantic web problem
Biomedical data are too complex
Is it really „semantic?“Will the computer understand, what does it
mean to be overweight?
Biomedical data as indicators
Phemomenological approach [Husserl]
“A thing is... properly an indication if and where it in fact serves to indicate something to some thinking being... a common circumstance [is] the fact, that certain objects or states of affairs of whose reality someone has actual knowledge indicate to him the reality of certain other objects or states of affairs, is the sense that his belief in the reality of the one is experienced (...) as motivating a belief or surmise in the reality of the other.” [Italics by E. H.].
Indicators
The reading of text „Height 190 cm“ evokes in the reader (indicates) quite tall person
Can be transformed Height 190 cm Taller than average
Transformations
Indicator ontology
Indicators can be grouped into „Indicator classes“ based on their common role in transformations (all recognizable indicators informing reader about body height)
Indicators can be transformed to other indicators manually or using computers
Indicator ontology
Next?
Ad-hoc ontology
Computer understandable syntactic information (module = relation)
Computer processable semantic information
Aposteriori constructed ontology
Next?
Service oriented architecture
Service oriented architectureAgents (computer or human)Emphasis on actionsAgent communication through message
exchange
A service is a set of actions that form a coherent whole from the point of view of service providers and service requesters
Web services architecture
Follows the SOA concepts
Emphasis on interoperabilityXML, XML, XMLProgramming model, implementation
technology, operating systém unaware
Distributed computing Physically distributed services and data Local (LAN) or large scale (WAN) Paralell computing Resource sharing
CPU time Peripherals – printers etc. Data
Application1. Large performance required2. Integration of resources
GRID
Large diversity of distributed (physical) resources
Access is globally unified through virtual layers (unifying abstractions)
ApplicationsComputational GridsScientific Data GridsVirtual Organizations Information and Knowledge Services
Conclusions
Documentation of biomedical algorithms is not yet well established Medigrid documentation service
Description of biomedical data in current classifications and ontologies is limited Indicator Ontology
Implementations are not reliably documented and therefore are not reusable GRID networking