p - MedUni Wien
Transcrição
p - MedUni Wien
Wissensbasierte Toxoplasmose-Diagnostik in der Schwangerschaft Klaus-Peter Adlassnig 1), Dieter Kopecky 1), Michael Hayde 2), Birgit Panzenböck 2), Arnold Pollak 2) 1) Institut für Medizinische Computerwissenschaften Abteilung für Medizinische Experten- und Wissensbasierte Systeme 2) Universitätsklinik für Kinder und Jugendheilkunde Klin. Abteilung für Neonatologie, angeborene Störungen und Intensivmedizin Medizinische Fakultät der Universität Wien Toxopert • Ziel: – Automatische Interpretation von Toxoplasmosetests im Zeitverlauf • Methode: – Integration des Toxopert in das Routine-EDV-System der jeweiligen medizinischen Einrichtung – Implementierung eines Entscheidungsgraphen – Befundausdruck mit Diagnose- und Therapievorschlag Oozysten von Toxoplasma gondii Aus dem Katzenkot. Natürliche Größe 12μm. (Aus: H. Aspöck, Toxoplasmose, Hoffmann-La Roche Wien, Wien 1992). Antibody Concentration after Toxoplasmosis Infection years years specific IgM detection months from: J.A.Pinard et al. (2003) Maternal Serologic Screening for Toxoplasmosis. Journal of Midwifery & Women’s Health 48, 308–316. Beispiel berechneter IgG-Trend Befund/ Untersuchung Datum SSW IgG Vorbefund 07.01.1999 10 negativ 1. 05.03.1999 18 negativ ⊥ 2. 30.04.1999 26 1:256 ↑↑ positiv 3. 27.05.1999 30 1:16 384 ↑↑ positiv IgM Beispiel Start neg Vorbefund pos keine Infektion 1. Untersuchung IgG≥1:4 und IgG≤1:256 IgG neg IgG neg 2. Untersuchung Infektion keine Infektion latente Infektion IgG≥1:4 und IgG≤1:256 IgG≥1:1024 Verdacht auf akute Infektion 3. Untersuchung ↓↓,↓,⊥,↑ latente Infektion oder widerspr. Befunde Akute Infektion ↑↑ akute Infektion IgG≥1:1024 akute Infektion TOXOPERT-I • Stand-alone Small Talk-Version TOXOPERT-Serologie Evaluierung von TOXOPERT-I Arzt → keine unzureichende oder Infektion inkonsistente Daten akut latent akut 17 0 0 0 17 latent 0 361 0 7 368 keine Infektion 0 0 606 0 606 unzureichende oder inkonsistente Daten 0 2 0 7 9 gesamt 17 363 606 14 1000 TOXOPERT-I ↓ gesamt TOXOPERT-II • Integrierte Laborversion Diskussion • Modellierung des medizinischen Wissens als Entscheidungsgraph – zeitliche Abfolge der Befunde implizit enthalten • Verlaufsbeurteilung bei Abweichung von Standardverläufen unzureichend – explizite Darstellung der Zeit erforderlich • Visualisierung des Inferenzpfades FuzzyTempTOXOPERT: Interpretation of Toxoplasmosis Serology Test Results Dieter Kopecky 1), Klaus-Peter Adlassnig 1), Michael Hayde 2), Andrea-Romana Prusa 2), and Birgit Panzenböck 2) Department of Medical Computer Sciences Section on Medical Expert and Knowledge-Based Systems University of Vienna Medical School, Spitalgasse 23, A-1090 Vienna, Austria 1) 2) Department of Pediatrics Division of Neonatology and Intensive Care University of Vienna Medical School, Währinger Gürtel 18–20 A-1090 Vienna, Austria General Considerations • medical problem – transplacental transmission of Toxoplasma gondii has to be treated immediately * question: Did an acute toxoplasmosis infection occur after conception? • objective – automatic interpretation of toxoplasmosis antibody test results in their course of time * IgG SFT, IgM ISAGA – automated generation of therapy proposals, if necessary • implementation – decision graph, nonmonotonic reasoning, temporal distances are modelled by fuzzy sets – integration into the routine laboratory information system of the toxoplasmosis laboratory – web-based interpretive system Decision Graph Start PF = ? T = [•] T = [•] 1.01 (I) T = [•] GA ≤ 29 ∧ IgM ∈ [−,±] ∧ IgG ∈[1: 4 ,1: 1024] 2.04 (N) T = [[2,4],•] GA ≤ 30 ∧ IgG = − 2.01 (N) T = [[2,4],•] T = [•] 3.08 (L) IgG ≥ 1: 4 ∧ [ TT ∈ ⇓, ↓, ⊥, ↑ ] 3.03 (A) IgG ∈ [1: 4,1:1024] ∧ IgM = + 3.08 (L) Latent (preconceptual) infection with Toxoplasma gondii is confirmed. The unborn is protected against an infection with Toxoplasma gondii. Additional serological control of this and future pregnancies is not necessary. T = [4,•] T = [4,•] GA ∈ [31,34] ∧ IgG = − 2.02 (N) T = [4,•] GA ≥ 35 ∧ IgG = − T = [4,•] T = [4,•] 2.03 (N) FuzzyTOXOPERT-III At Least 3 to 5 Weeks ⎧ 0,..................∀t , t ∈ [0,3) ⎪t ⎪ μ temp (t ) = ⎨ − 1.5,..........∀t , t ∈ [3,5] ⎪2 ⎪⎩1,...................∀t , t ∈ (5,+∞ ] temporal compatibility μ 1.00 0.75 0.50 0.25 0.00 1 2 3 4 5 6 7 8 9 10 11 12 t in weeks Within 7 Weeks ⎧1,....................∀t , t ∈ [0,7] ⎪ t ⎪ μ temp (t ) = ⎨− + 4.5........∀t , t ∈ (7,9 ) ⎪ 2 ⎪⎩0,...................∀t , t ∈ [9,+∞ ) temporal compatibility μ 1.00 0.75 0.50 0.25 0.00 1 2 3 4 5 6 7 8 9 10 11 12 t in weeks Knowledge Base Taking Temporal Compatibility into Account–I Interpretation 3.08 [0.0-0.3] Latent (preconceptual) infection with Toxoplasma gondii is very probable. [0.3-0.8] Latent (preconceptual) infection with Toxoplasma gondii is rather certain. [0.8-1.0] Latent (preconceptual) infection with Toxoplasma gondii is confirmed. The unborn is protected against an infection with Toxoplasma gondii. ]0.0-0.3] Serological followup required at 1-2 weeks. [0.3-0.8] Serological followup required at 2-3 weeks. [0.8-1.0] Additional serological control of this and future pregnancies is not necessary. ]0.0-0.99] Since the obligate time interval of @1 weeks has not been kept between the two tests on @2 and @3, it is possible, that a probable titer rise has not been recognized. IF Graph Start 1.01 2.05 3.08 Prel.Find. ? Findings GA IgG IgG trend 1 2 3-i [30,37] [1:4,1:1024] ≥ 1:4 ≥ 1:4 ⇓↓⊥↑ ⇓↓⊥↑ IgM ±+ Time • • [[2, 4], •] • Knowledge Base Taking Temporal Compatibility into Account–II OR Graph Start 1.01 2.04 3.08 OR Graph Start 1.02 2.05 3.08 OR Graph Start 1.02 2.15 3.08 OR Graph Start 1.02 2.04 3.08 Prel.Find. ? Prel.Find. – Prel.Find. – Prel.Find. – Findings GA IgG IgG trend 1 2 3-i ≤ 29 [1:4,1:1024] ≥ 1:4 ≥ 1:4 ⇓↓⊥↑ ⇓↓⊥↑ Findings GA IgG IgG trend 1 2 3-i [30,37] [1:4,1:1024] ≥ 1:4 ≥ 1:4 ⇓↓⊥↑ ⇓↓⊥↑ Findings GA IgG IgG trend 1 2 3-i ≤ 37 [1:4,1:1024] ≥ 1:4 ≥ 1:4 ⇓↓⊥↑ ⇓↓⊥↑ Findings GA IgG IgG trend 1 2 3-i ≤ 29 [1:4,1:1024] ≥ 1:4 ≥ 1:4 ⇓↓⊥↑ ⇓↓⊥↑ IgM ±+ IgM –± IgM + IgM –± Time • • [[2, 4], •] • Time • • [[2, 4], •] • Time • • [[2, 4], •] • Time • • [[2, 4], •] • Diagnostic interpretation, therapy recommendation, and proposals for further examinations for a woman with acute toxoplasmosis. The second inference cycle tried to eliminate test results that have lowered the overall temporal compatibility. Discussion on FuzzyToxopert • modeling of temporal concepts by fuzzy sets • degrees of applicability of rules < 1 (attributed to decision paths) result in soft (= less decisive) interpretations Logikbasierte Repräsentation CADIAG: Computer-Assisted Diagnosis • predecessor (1968–1974): – propositional logic: hepatology, rheumatology • CADIAG-I (1976–1983, 1990–1992): – three-valued logic, predicate logic: rheumatology, gastroenterology, hepatology • CADIAG-II (1978–present): – fuzzy set theory, fuzzy logic, compositional rule of inference with occurrence and confirmability (numerical values): rheumatology, gastroenterology, hepatology, neurology, nosocomial infections) • CADIAG-III (1992–1994): – extension of fuzzy operators (missing values in medicine), fix point inference • MedFrame/CADIAG-IV (1986, 1994–present): – extended data-to-symbol conversion, inference with SDoccurrence, SnotD-occurrence, SD-confirmability, SnotDconfirmability (linguistic values), generalization to symptoms, diseases, and therapies Computer-Assisted Medical Diagnosis and Therapy • “Reasoning Foundations of Medical Diagnosis” (1959) in Science by Ledley and Lusted – symbolic logic ∗ symptom complexes = logic combinations; diagnoses implicated or excluded – probability ∗ frequency of symptoms with diseases, frequency of symptoms and frequency of diseases in a population; most probable diagnosis – value theory ∗ decision trees; optimal treatment Relationships in CADIAG-I positive association necessity sufficiency – O: obligatory occurring – F: facultative occurring – C: confirming – N: not confirming OC: obligatory occurring and confirming FC: facultative occurring and confirming ON: obligatory occurring and not confirming FN: facultative occurring and not confirming exclusion EX: excluding Medizinische Relationen in Prädikatenlogik • Relation OB (obligatorisch auftretend und beweisend): 1. alle p mit S haben D 2. alle p mit D haben S 3. es gibt mindestens ein p mit S ∧ S OB D = ∀p[S(p)→D(p)] ∧ ∀p[D(p)→S(p)] ∧ ∃p[S(p)] • Relation FB (fakultativ auftretend und beweisend): 1. alle p mit S haben D 2. nicht alle p mit D haben S 3. es gibt mindestens ein p mit S ∧ S FB D = ∀p[S(p)→D(p)] ∧ ∃p[D(p)∧¬S(p)] ∧ ∃p[S(p)] • Relation ON (obligatorisch auftretend aber nicht beweisend): 1. alle p mit D haben S 2. nicht alle p mit S haben D 3. es gibt mindestens ein p mit D ∧ S ON D = ∀p[D(p)→S(p)] ∧ ∃p[S(p)∧¬D(p)] ∧ ∃p[D(p)] • Relation FN (fakultativ auftretend und nicht beweisend): 1. nicht alle p mit S haben D 2. nicht alle p mit D haben S 3. es gibt mindestens ein p mit S und D ∧ S FN D = ∃p[S(p)∧¬D(p)] ∧ ∃p[D(p)∧¬S(p)] ∧ ∃p[S(p)∧D(p)] • Relation A (ausschließend) 1. alle p mit S schließen D aus 2. es gibt mindestens ein p mit S und nicht D 3. es gibt mindestens ein p mit D und nicht S ∧ S A D = ∀p[S(p)→¬D(p)] ∧ ∃p[S(p)∧¬D(p)] ∧ ∃p[D(p)∧¬S(p)] Relationships in CADIAG-II positive association necessity ∈ (0,1] sufficiency ∈ (0,1] exclusion necessity = [0] sufficiency = [0] Logikmodule IF. . . an order for DIGOXIN is placed Pharmacist enters medication order The system works behind the scenes as an “active agents” evaluating clinical scenarios. AND The patient is already receiving FUROSEMIDE System checks patient’s medication System checks patient’s lab results CONDITIONS MET “fi res ” The patient’s POTASSIUM LEVEL is low The system immediately takes the following predefined actions: Notify the ordering pharmacist of the drug interaction and contact the requesting physician immediately AND rul e AND THEN. . . Notify the clinical pharmacist for follow-up and documentation AND Notify the requesting physician of the situation and suggest a potassium supplement Drug-Laboratory Triggers Clinical Pathways and Guidelines Enacting Guidelines FuzzyKBWean: A Fuzzy Control System for Weaning from Artificial Ventilation C. Schuh1, M. Hiesmayr2, K.-P. Adlassnig1, M. Kolb2 1Department of Medical Computer Sciences 2Department of Cardiothoracic and Vascular Anaesthesia and Intensive Care University of Vienna Medical School Objective • mechanically ventilated patients after cardiothoracic surgery in an intensive care unit (ICU) • proposals for changes in ventilator settings during the three phases of mechanical ventilation (stabilization, weaning, and finally extubation of the patient) • open-loop system: integration into the patient data management system (PDMS); time resolution of 1 minute • closed-loop system as a long-term objective: integration into the ventilator (auto-mode) Motivation We found that nurses and respiratory therapists, using protocol guidance, weaned patients from mechanical ventilation safely and more quickly than the team following the traditional practice of physician-directed weaning. Koleff MH, et al : A Randomized, Controlled Trial of Protocoll-Directed versus Physician-Directed Weaning from Mechanical Ventilation. Critical Care Medicine 1997; 25, 567–574. Structure of FuzzyKBWean Methods • phase-dependent fuzzy sets • linguistic if/then rules – if: patient’s physiological parameters and ventilator measurement parameters (in a defined context) – then: proposals for changes in ventilator settings • fuzzification step – arithmetic, statistical, comparative, logical, temporal, and control operators • defuzzification step – Sugeno’s center of gravity method • verification by the attending physician, i.e., open-loop Patient’s Physiological Parameters • oxygenation O2 – arterial oxygen partial pressure pO2 (directly) – oxygen saturation SpO2 (indirectly: pulsoxymetry) • ventilation: CO2-elimination (~ to alveolar ventilation) – arterial carbondioxide partial pressure pCO2 (directly) endtidal carbondioxide EtCO2 (indirectly) • expired tidal volume Tve • respiratory rate Vrate Fuzzy Control inference process with linguistic fuzzy rules defuzzification with Sugeno’s center of gravity method fuzzification with fuzzy sets PaO2 [mmHg] PaCO2 [mmHg] measurement and observational level FiO2-change [%] Knowledge Base fuzzy sets PaO2 1 low very low 0 70 PaCO2 low 1 very low 0 35 90 95 100 [mmHg] 40 45 50 very high 55 linguistic fuzzy rules rule 1: If PaO2 = very low and PaCO2 = low, then FiO2-change = +10 rule 2: If PaO2 = normal and PaCO2 = normal, then FiO2-change = –5 rule 3: If PaO2 = high and PaCO2 = normal, then FiO2-change = –10 etc. 85 high high normal 30 75 normal [mmHg] PATIENT: PaO2 = 96 mmHg, PaCO2 = 42 mmHg fuzzification PaO2 1 0.8 low very low normal high 0.2 70 PaCO2 low 1 very low 0 35 85 high 40 45 50 42 90 95 100 [mmHg] 96 very high normal 30 75 55 inference process rule 2: If PaO2 = normal and PaCO2 = normal, then FiO2-change –5 rule 3: If PaO2 = high and PaCO2 = normal, then FiO2-change –10 rule 2: min (PaO2 = normal, PaCO2 = normal) = min (0.8, 1) = 0.8 rule 3: min (PaO2 = high, PaCO2 = normal) = min (0.2, 1) = 0.2 defuzzification FiO2-change = 0.8*(–5)+0.2*(–10) = –6 0.8+0.2 [mmHg] Results–I • 23 variables – 74 fuzzy sets (phase-dependent) • 16 if/then rules – 4 rules checking for measurement errors and validity – 3 rules for ventilation (normal range, hypoventilation, hyperventilation) – 4 rules for oxygenation (stabilization, oxygenation normal, hypoxia, severe hypoxia) – 4 rules for intermediate states (increased EtCO2, decreased EtCO2, phase changes) – 1 rule for extubation Results–II • 10 prospectively randomized patients – FuzzyKBWean reacted correctly 131 (SEM 47) minutes earlier than the attending physician – adjustment of ventilation parameters was more reliable than adjustment of oxygenation (EtCO2 is more reliable as SpO2) – phase-specific rules often proposed too small changes of the ventilator settings • temporal rule blocking, fuzzy set adaptations, employing thresholds to avoid oscillations Results–III Delay of Staff Reaction in Case of Hyperventilation patient episode proposed change at effective change at delay (min) M 1 2 3 1 1 1 2 1 2 1 2 1 2 3 4 1 16:30 23:50 04:27 21:17 22:57 22:30 01:15 13:45 20:20 17:03 08:48 20:02 20:12 20:30 22:47 21:05 19:42 01:55 07:43 03:58 23:16 01:15 01:50 14:45 20:48 18:27 16:36 20:12 20:30 22:47 23:22 21:40 192 125 196 401 19 165 35 60 28 104 468 10 18 137 35 35 B D E K C G K Discussion on FuzzyKBWean • methodology – minimal number of therapeutically significant classes per variable – gradual transition between variable classes • adequate consideration of the inherent fuzziness of medical concepts – intuitive if/then rules at the knowledge level • physician’s medical knowledge was transferred to FuzzyKBWean • clinical trial – periods of deviation from the target parameters are shorter • contribution to the patient’s safety and comfort – closed-loop: detection of artifacts and information obtained by direct observation of the patient FuzzyARDS: Knowledge-Based Monitoring and Decision Support F. Steimann1, H. Steltzer2, K.-P. Adlassnig1 1Department of Medical Computer Sciences Section on Medical Expert and Knowledge-Based Systems University of Vienna Medical School, Austria 2 Department of Anesthesiology and General Intensive Care Medicine University of Vienna Medical School, Austria Objective • knowledge-based decision support – monitoring patients with acute respiratory distress syndrome (ARDS) – early detection of ARDS – therapy advice in ARDS cases • international study (Vienna, Berlin, Marburg, Paris, Milan) – to improve ARDS definition – to compare therapy entry criteria Wir beobachteten, daß Krankenschwestern und Therapeuten, die für die Entwöhnung der Patienten von der künstlichen Beatmung Protokolle verwendeten, diese schneller und sicherer entwöhnten, als das Team, welches traditionellerweise den Anordnungen der Ärzte folgte. Koleff MH, et al. : A Randomized, Controlled Trial of Protocoll-Directed versus Physician-Directed Weaning from Mechanical Ventilation. Critical Care Medicine 1997; 25, 567–574. VM Ventilator Manager: Stanford University Medical School, 1979 • experimental, rule-based expert system for on-line data interpretation at the ICU STATUS-RULE: STABLE HEMODYNAMICS DEFINITION: defines stable hemodynamics based on blood pressure and heart rate APPLIES TO: patients on volume, cmv, assist, T-piece COMMENT: look at mean arterial pressure for changes in blood pressure and systolic blood pressure for maximum pressures IF 1) heart rate is acceptable, and 2) pulse rate does not change by 20 beats/minute in 15 minutes, and 3) mean arterial pressure is acceptable, and 4) mean arterial pressure does not change by 15 torr in 15 minutes, and 5) systolic blood pressure is acceptable, THEN the hemodynamics are stable. Example of a heuristic VM-rule. First Approach: Method • if/then rule for the diagnosis of ARDS under consideration of risk factors and first clinical signs • if/then rule for the diagnosis of manifest ARDS, based on Murray and Morel scores ARDS: Rule for Early Detection risk = .or. .or. .or. .or. .or. .or. .or. .or. .or. .or. .or. .or. SEPSIS; TRAUMA; ASPIRATION; PNEUMONIA; SHOCK; TRANSFUSION; INHALATION; PANKREATITIS; DROWNING; FAT EMBOLISM; COAGULOPATHY; BURNING; KIDNEY FAILURE .or. .or. .or. .or. .or. .or. .or. PAO2 < 75 .and. FIO2 ≥ 0.5; PAO2 ≤ 65 .and. FIO2 ≥ 0.4; PAO2 < 250 .and. FIO2 = 1.0; PAO2 / FIO2 < 250; ROENTSTAD ≥ 1; SHUNT > 20; TOTRAUMVEN; COMPLIANCE < 50 signs = signs = signs .and. PCWP < 18 ARDS: rule for early detection = risk .and. signs First Approach: Retrospective Tests • • 1,104 data records from 32 patients suffering from ARDS 116 data records from 8 patients not suffering from ARDS results in ARDS diagnosis: • • sensitivity: 89.6% specificity: 18.1% results in manifest ARDS: • • sensitivity: 68.8% specificity: 69.8% • diagnosis in accordance with Murray and Morel scores: 69.7% (all 1,220 data records) First Approach: Problems • ARDS is a fuzzily defined nosological entity ⇒ a crisp definition of ARDS is inadequate ⇒ in particular, the commitment to crisp limits of findings is unintuitive (see figure) ⇒ thresholds of PaO2 and FiO2 for patients suffering from ARDS: PaO2 FiO2 ⇒ hypothesis: it is practically impossible to characterize ARDS by a crisp definition Second Approach: Methods • development of a state transition diagram (deterministic automaton) – – mutually exclusive states (no ARDS, suspected ARDS, early detection of ARDS, confirmed ARDS, manifest ARDS, severe ARDS, ...) crisp transition conditions (PaO2/FiO2 < 250, ...) monitoring chest radiograph, spontaneous breathing, artificial respiration, and circulation negative PaO2/FiO2 < 250 for 1 day no risk factors suspected ARDS risk factors exclusion criteria not clarified PaO2/FiO2 no ARDS confirmed ARDS manifest ARDS CT: pulmonary lesion < 1/3 and Morel score < 2.5 and Murray score < 2.5 or ... ARDS diagnosis risk factors no risk factors CT: pulmonary lesion < 2/3 and Morel score > 2.5 and Murray score > 2.5 or ... severe ARDS chest radiograph, spontaneous breathing, artificial respiration, or circulation pathological CT: pulmonary lesion > 2/3 and Morel score = 4 and Murray score > 3.5 ECCO2/ ECMO Weighted Scores Scoring System definition of two criteria: PaO2/FiO2 < 150 (with weight 2) shunt > 30 (with weight 1) case 1: criteria PaO2/FiO2 = 100 shunt = 25 score fulfillment 2/(2+1) 0/(2+1) 2/3 case 2: criteria PaO2/FiO2 = 160 shunt = 35 score fulfillment 0/(2+1) 1/(2+1) 1/3 New Approach: Methods • development of a fuzzy state transition diagram (fuzzy automaton) – with mutually non exclusive states (gradual transition from one state to another, concurrent partial presence of several states; the clinician is able to consider the situation carefully) – with fuzzy conditions for the transition of states (fuzzy medical concepts, fuzzy trend detection) monitoring chest radiograph, spontaneous breathing, artificial respiration, and circulation negative PaO2/FiO2 < 250 for 1 day no risk factors suspected ARDS risk factors exclusion criteria not clarified PaO2/FiO2 no ARDS no risk factors ARDS confirmed manifest ARDS CT: pulmonary lesion < 1/3 and Morel score < 2.5 and Murray score < 2.5 or ... ARDS diagnosis risk factors CT: pulmonary lesion < 2/3 and Morel score > 2.5 and Murray score > 2.5 or ... severe ARDS chest radiograph, spontaneous breathing, artificial respiration, or circulation pathological CT: pulmonary lesion > 2/3 and Morel score = 4 and Murray score > 3.5 ECCO2/ ECMO Scoring System with Fuzzy Concepts Scoring System with Fuzzy Concepts PaO2 / FiO2 < 150 (200) with weight 2 Shunt > 30 (20) with weight 1 criteria PaO2/FiO2 = 164 shunt = 23 score } 2+1 = 3 score and fulfillment 0.72 * (2/3) = 0.72 * 2/3 0.3 * (1/3) = 0.3 * 1/3 0.58 degree degree 1 1 0.72 0.3 0 150 164 200 PaO2/FiO2 0 20 23 30 shunt Patient Conditions Condition Definition with “fuzzified by” adequate oxygenation SaO2 above 97% (93%) for 5 minutes hypoxemia SaO2 between 90% and 93% (87% and 97%) for 2 minutes high FiO2 FiO2 above 60% for 30 seconds low FiO2 FiO2 below 60% for 30 seconds rapidly improving oxygenation SaO2 increasing from 87–95% (85–99%) to 97–100% (93–100%) within 30–90 seconds (5–120 seconds) slowly decreasing oxygenation SaO2 above 96% (91%) steady or decreasing to 94% (89%) within 25 minutes Graphical representation Trend Detection Applied to SaO2 (in part filtered through moving average) 12 hour period State Transition Diagram of the Automaton start normal improved after hand bagging responding to high FiO2 hypoxic not improved after hand bagging not responding to high FiO2 Definition of States under Consideration State Interpretation start normal initial state, undecided oxygenation is satisfactory without additional effort such as increased FiO2 oxygenation is too low and should be improved high FiO2 has affected oxygenation positively high FiO2 does not have the desired effect hand bagging has persistently improved oxygenation hypoxic responding to high FiO2 not responding to high FiO2 improved after hand bagging not improved after hand bagging hand bagging shows no satisfactory effect Linguistic State Transitions From state On condition To state start adequate oxygenation hypoxemia normal hypoxic normal hypoxemia hypoxic hypoxic low FiO2 ∧ adequate oxygenation high FiO2 ∧ rapidly improving oxygenation high FiO2 ∧ hypoxemia normal responding to high FiO2 not responding to high FiO2 low FiO2 ∧ slowly decreasing oxygenation low FiO2 ∧ hypoxemia improved after hand bagging not responding to high FiO2 low FiO2 ∧ hypoxemia high FiO2 ∧ adequate oxygenation hypoxic responding to high FiO2 improved after hand bagging adequate oxygenation hypoxemia normal hypoxic not improved after hand bagging hypoxemia hypoxic responding to high FiO2 not improved after hand bagging Distribution of States over Time Discussion on FuzzyARDS • concept modeling and trend detection with fuzzy sets – inherent fuzziness of medical and temporal concepts, unsharpness of boundaries • monitoring with fuzzy automata – fuzzily defined nosological concepts (partial lack of medical theory) • visualization on the ICU monitor – integration in the PDMS – dynamic visualization of changes of states – display of state transitions over time