Outline. Asbru. Asbru Language. AsbruView. Motivation. The Asgaard/Asbru Project. Evaluation, Conclusion

Transcription

1 Asgaard / Asbru Project : Designing Task-Specific Problem-Solving Methods to Support the Design and the Execution of -Oriented Skeletal Plans Peter Johnson Silvia Miksch Yuval Shahar Stanford University Stanford Medical Informatics (SMI) International Cooperation Vrije Universiteit Amsterdam University of Aberdeen University of Augsburg CBO, Dutch - Healthcare Improvement Klaus Hammermüller Robert Kosara Andreas Seyfang and students Vienna University of Technology Vienna, Austria Yuval Shahar and students Ben-Gurion University of the Negev Beer-Sheva, Israel Outline Motivation Medical Therapy Planning Plan Representations Limitations The Asgaard/Asbru Project Asbru Language Terminology Knowledge Roles AsbruView Metaphor Graphics Evaluation, Conclusion New Language Asbru

2 Medical Therapy Planning Clinical Protocols Current Representations of Protocols Free Text Tables Flow-Charts Acute Severe Asthma Guideline Indications: - Cannot complete sentences - Pulse > 110 beats/minute - Respiration > 25 breaths/minute - Peak flow < 50% of predicted or best Therapy: Oxygen 40-60% if available Nebulized salbutamol 5mg or nebulized terbutaline 10mg Oral prednisolone mg or i/v hydrocortisone 200mg Monitor responses minutes after nebulisation If any signs of acute asthma persist: +Arrange hospital admission +While awaiting ambulance: repeat nebulised beta 2 and give with nebulised ipratropium 500 micrograms OR give subcutaneous terbutaline OR give slow i/v aminophylline 250mg Else if symptoms have improved, step up usual treatment; follow up Medical Therapy Planning Clinical Protocols Current Representations of Protocols Free Text Tables Flow-Charts Used for Communication Quality Assessment

3 Plan Representation requirements Plan Representation missing requirements Huge volume of data Pre- and post-conditions are needed to control the execution of durative actions or plans Goal may not be achievable in time Sequential, parallel, and cyclical execution of plans is necessary Large domain knowledge is available Incomplete and non-deterministic information Planning agent is not omniscient Unobservable underlying processes Goals are not categorical and unchanged over time Everything is durative, not instantaneous Multiple time lines based on different granularities Plans can be suspended Interleaving of plan design and execution need Asgaard / Asbru Concept Asbru s Key Features INPUT -Oriented Clinical Guidelines Patient Data Task-Specific PSM Abstraction Visualization Sharable Plan-Specification Library OUTPUT -Oriented Verification Validation Explanations Plan Selection - Monitoring Critiquing Plan Execution Future Projection Evaluation Hierarchical Decomposition of Plans Temporal Annotations & Uncertainty Knowledge Roles Preferences Intentions Conditions Effects Plan Layouts & TASKS

4 Hierarchical Decomposition of Plans Parameter Proposition Plans A B C E PLAN AA PLAN A1 D E E F H PLAN A2 G I E Parameter or its Abstraction Value Context Annotation Example: Temperature in the context of controlled ventilation is high for more than 3 hours starting from 24 to 26 hours after I-RDS was diagnosed. (Temperature High Controlled-Ventilation [[24 HOURS, 26 HOURS], [_, _], [3 HOURS,_], I-RDS-diagnosed]) Asbru 's Annotation Asbru s Knowledge Roles (1/2) Definition: [[ESS, LSS], [EFS, LFS], [MinDu, MaxDu], Reference] MinDu Preferences strategy... Effects Arg-dependency Overall MaxDu Reference ESS LSS EFS LFS ESS LSS EFS LFS Example: [[_, _], [_, _], [180 MIN, _], *self*] Earliest Starting Shift Latest Starting Shift Earliest Finishing Shift Latest Finishing Shift 180 Min undef. MinDu MaxDu Reference Point Minimum Duration Maximum Duration *self* undef. undef. undef. undef. Intermediate state Intermediate action Intentions Overall state pattern Overall action pattern Conditions... Plan body Sequential Concurrent Cyclical Parallel Any Order Unorder

5 Asbru s Knowledge Roles (2/2) Filter Setup Suspend Conditions Abort Complete Activate Reactivate I-RDS Example in Asbru 6.5 (PLAN controlled-ventilation (PREFERENCES (SELECT-METHOD BEST-FIT)) (INTENTION:INTERMEDIATE-STATE (MAINTAIN STATE(BG) NORMAL controlled-ventilation *)) (INTENTION:INTERMEDIATE-ACTION (MAINTAIN STATE(RESPIRATOR-SETTING) LOW controlledventilation *)) (SETUP-PRECONDITIONS (PIP (<= 30) I-RDS *now*) (BG available I-RDS [[_, _], [_, _],[1 MIN,_](ACTIVATED initial-phase#1)])) (ACTIVATED-CONDITIONS AUTOMATIC) (ABORT-CONDITIONS ACTIVATED (OR (PIP (> 30) controlled-ventilation [[_, _], [_, _], [30 SEC, _], *self*]) (RATE(BG) TOO-STEEP controlled-ventilation [[_, _], [_, _], [180 MIN,_], *self*]))) (SAMPLING-FREQUENCY 10 SEC)) (COMPLETE-CONDITIONS (FiO2 (<= 50) controlled-ventilation [[_, _], [_, _], [180 MIN, _], *self*]) (PIP (<= 23) controlled-ventilation [[_, _], [_, _], [180 MIN, _], *self*]) (f (<= 60) controlled-ventilation [[_, _], [_, _], [180 MIN, _], *self*]) (patient (NOT DYSPNEIC) controlled-ventilation [[_, _], [_, _], [180 MIN, ], *self*])) (STATE(BG) (OR NORMAL ABOVE-NORMAL) controlled-ventilation [[_, _], [_, _], [180 MIN,_], *self*]) (SAMPLING-FREQUENCY 10 MIN)) (DO-ALL-SEQUENTIALLY (one-of-increase-decrease-ventilation) (observing))) I-RDS Example in Asbru 7.1d <xml version="1.0" encoding="utf-8" > <!DOCTYPE plan-library (View Source for full doctype...)> - <plan-library> - <domain-defs> - <domain name="i-rds"> - <record-def name="patient"> <field-def name="name" type="string" /> <field-def name="sex" type="maleorfemale" /> </record-def> - <qual-scale-def name="rate-change"> <qual-entry entry="normal" /> <qual-entry entry="too-steep" /> </qual-scale-def> - <qual-scale-def name="bg_state"> <qual-entry entry="normal" /> <qual-entry entry="strange" /> <qual-entry entry="critical" /> </qual-scale-def> - <qual-scale-def name="respirator-setting"> <qual-entry entry="low" /> <qual-entry entry="normal" /> <qual-entry entry="high" /> </qual-scale-def> - <qual-scale-def name="maleorfemale"> <qual-entry entry="male" /> <qual-entry entry="female" /> </qual-scale-def> - <qual-scale-def name="facecolor"> <qual-entry entry="not_dypneic" /> <qual-entry entry="dyspneic" /> </qual-scale-def> - <unit-def name="frequency" rounding="nearest"> - <compound-def> <numerator /> - <denominator> <unit-class name="time" exponent="1" /> </denominator> </compound-def> </unit-def> - <unit-def name="flow-unit" rounding="nearest"> - <compound-def> - <numerator> <unit-class exponent="1" name="amount" /> </numerator> <denominator /> </compound-def> </unit-def> - <parameter-def name="bg" required="yes" type="bg_state"> - <qual-parameter-def scale="bg_state"> - <limits> <context-ref name="controlled-ventilation" /> <limit-entry value="1" /> <limit-entry value="5" /> <limit-entry value="7" /> <limit-entry value="9" /> <numerical-constant unit="s" value="30" /> </limits> <raw-data-def mode="manual" /> </qual-parameter-def> <numerical-constant unit="s" value="10" /> </parameter-def> - <parameter-def name="repirator-setting" required="no" type="respirator-setting"> - <qual-parameter-def scale="respirator-setting"> - <limits> <context-ref name="i-rds" /> <limit-entry value="low" /> <limit-entry value="normal" /> <limit-entry value="high" /> <numerical-constant unit="s" value="10" /> </limits> - <spread-def> <numerical-constant unit="s" value="10" /> <numerical-constant unit="s" value="10" /> <raw-data-def mode="automatic" /> </spread-def> </qual-parameter-def> <numerical-constant unit="s" value="10" /> </parameter-def> - <parameter-def name="pip" required="yes" type="pressure"> <raw-data-def mode="automatic" /> <numerical-constant unit="s" value="10" /> </parameter-def> - <parameter-def name="f" required="yes" type="frequency"> <raw-data-def mode="automatic" /> <numerical-constant unit="s" value="30" /> </parameter-def> - <parameter-def name="fio2" required="yes" type="flow-unit"> <raw-data-def mode="automatic" /> <numerical-constant unit="s" value="10" /> </parameter-def> - <parameter-def name="facecolor" required="yes" type="facecolor"> <raw-data-def mode="manual" /> <numerical-constant unit="h" value="1" /> </parameter-def> </domain> </domain-defs> <value-defs /> - <plans> - <group name="controlled-ventilation"> -... Impossible to use for Domain Experts

6 Visualization Challenges AsbruView Hierarchical Decomposition Conditions Plan Types Temporal Order Compulsory vs. Optional Plans Cyclical Plans Temporal Uncertainty Traffic Interface to the Plan-Representation Language Asbru Track 1 Track 2

7 AsbruView s Dimensions Anatomy of a Plan 2. Parallel Plans 3. Subplan Levels Subplan AA Subplan AB Plan A 4. Color Levels Filter Precondition Complete Condition Abort Condition Suspend Condition Reactivate Condition 1. Flow of Parallel Plans Setup Precondition AsbruView vs. Flow-Charts Levels Sequential Plans Topological View Plan A Parallel Plans Y Setup- Precondition Y N Y Fulfill Filter- Precondition Fulfillable N N Find other plan Y N Abort Y daction Plan Completed Complete N Y Suspend N Y Abort plan Reactivate N Y Abort N Temporal View

8 Parallel Plans Some-Any-Order Plans Topological View Topological View Temporal View Temporal View Cyclical Plans Level of Detail Topological View Open Close Temporal View

9 AsbruView AsbruView - SopoView Evaluation Asbru s Usability Metaphors Colors Two Views Annotations Speed (Plan Placement) Transfer of Knowledge Verification and Validation Evaluation of Protocols Knowledge Roles Temporal Dimensions Reuse Quality Assurance

10 Conclusion Overview: Research Projects Asbru Clinical Protocols Asbru not suitable for Physicians AsbruView Graphical Representation Metaphors Glyphs Two Views VIE-PNN Asgaard/Asbru VIE-Nmed VIE-VENT Asgaard/SopoView Pulsoximetry Patient Advocate INPUT -Oriented V & V Plan Verification Protocols Clinical Semantic Depth of Field (SDOF) Data Validation Patient Data + * ** + VIE-Nmed

11 VIE-VENT SopoView Patient Advocate Pulsoximetry

12 VuV Building Blocks -Oriented Skeletal Planning Information Visualization Plan Verification Data Validation Temporal Data Abstraction INPUT -Oriented Clinical Protocols Patient Data * * * + + INPUT -Oriented Plan Verification Clinical Protocols Data Validation Patient Data + * + * * SDOF Principle Idea: 2D SDOF Scene Spatial Arrangement Data 2D 3D Viewing and Camera Model irrelevant focus plane Relevance and Blurring Photorealistic Adaptive... Selection Distance... r [0; 1] b [1; [ relevant

13 Focus and Context: Threats Focus, but no Context... Focus and Context: Threats Focus and Context: Covers

14 Questions Incomes Travels Gewerkschaftspolitik: innerbetrieb. Erhöhung: FAHRTROUTEN A1: Wechsel zwischen Verwendungsgruppen und Wechsel der Firma Monatliches Einkommen in den Jahren 1-25 % Erhöhg. KV-Abschlüsse: 1,15% IST-Abschlüsse: 0,58% normal Theodor Tandler Techniker Er hat die HTL besucht. Er hat Berufserfahrung. Alter: 23 Jahre normal keine Umrhg. End-Umrhg. +Wechsel -Wechsel Summe des Einkommens nach... Jahren Annahmen... ohne gepl.... mit "normale"... keine... mit erfolgreichem für die "normale" Umrhg., aber Fa.wechsel Einkommensentwicklung Umreihung Umrhg. am nach Ar- Einkommensentwicklung in höhere VG Fa.wechsel Ende der VG beitslosigkeit normal keine Umrhg. End-Umrhg. +Wechsel -Wechsel Einstufung zu Beginn 5 Jahre des Arbeitsverhältnisses III 4 10 Jahre Überzahlung zu Beginn 22% 15 Jahre Umreihung in höhere Verwen- 20 Jahre dungsgruppe geplant nach: III 8 25 Jahre Max. Zahl VG-Jahre i. d. VG 18 Bruttogehälter (14mal pro Jahr); Rundung auf 1000,-- ÖS; inflationsbereinigt Letzter EB-Sprung i. d. VG bei 16