Watson in Healthcare & Lifesciences Op weg naar Cognitive Computing

Watson in Healthcare & Lifesciences Op weg naar Cognitive Computing

27 Januari 2014

To wrest from nature the secrets which have perplexed philosophers in all ages, to track to their sources the causes of disease, to correlate the vast stores of knowledge, that they are quickly available for the prevention and cure of disease - these are our ambitions Sir William Osler (1849-1919) Co-founder of the Faculty of Medicine Johns Hopkins University

© 2014 International Business Machines Corporation

2

Introductie spreker Mijn naam: Nicky Hekster Technical Leader Healthcare & LifeSciences IBM Actief in ICT sinds 1987, in HC & LS sinds 2006 Leveranciersvoorzitter IHE Nederland [email protected] Te vinden op

3


Agenda De context van de zorg Watson, wat is er tot nu toe gedaan? Technisch fundament Voorbeelden van Watson Advisors Wanneer is iets Watson waardig? Thomas J. Watson (1874 – 1956)

Voorbeelden van de “route naar Watson” Discussie over hoe verder


4

14 februari 2011

5


Inconsistente kwaliteit en toenemende kosten vragen om verandering • Medicatiefouten leiden jaarlijks tot zo'n 90.000 onnodige opnames

Verhoging van doeltreffendheid en doelmatigheid

Verbetering van de klinische bedrijfsvoering

6


• De kosten van incidenten in de zorg bedragen jaarlijks 4 miljard euro • Geneesmiddelen voor kanker en Alzheimer werken in 75 -80% van de gevallen niet

• Het aantal mensen dat in Nederland per jaar overlijdt aan vermijdbare medische fouten is ca. 1500 tot 1600 • Er gaan jaarlijks miljoenen verloren door administratieve en klinische verspilling, fraude, en misbruik

6

Zorgdata – hoge volumes en enorme variatie Geleverde PetaBytes

EPD Radiologie Cardiologie Pathologie MDL Dermatologie

500 MB/4D beeld

Gestructureerd

1 MB/2D beeld 2006

2007

2008

2009

2010

2011 2004

>2009

Ongestructureerd

Ongeveer 80% van alle opgeslagen zorgdata is ongestructureerd1

Zorgdataopslagcapacit eit groeit met 35% per jaar2

30% van de wereldwijde gegevensopslag bestaat uit medische beelden3

Data jaarlijks uitgewisseld tussen zorginformatiesystemen: In 2010: 283 terabyte, en in 2020: 78 petabyte - 774 miljoen verbonden apparaten4 1AIIM

website, geaccepteerd percentage studie door de Enterprise Strategy Group

2Recente


Medical Transcription Discharge Summary Sample # 2:

3IBM

Global Technology Outlook for 2005

4http://www.machinaresearch.com/healthcare2020.h

tml

7

Cardiology Consultation Transcribed Medical Transcription Sample Reports

REFERRING PHYSICIAN: John Doe, MD CONSULTING PHYSICIAN: Jane Doe, MD HISTORY OF PRESENT ILLNESS: This (XX)-year-old lady is seen in consultation for Dr. John Doe. She has been under consideration for ventral hernia repair and has a background of aortic valve replacement and known coronary artery disease. The patient was admitted with complaints of abdominal pain, anorexia, and vomiting. She underwent a CT scan of the abdomen and pelvis ADMITTING DIAGNOSIS: Syncope. and this showed the ventral hernia involving the transverse colon, but without strangulation. There was an atrophic right kidney. She had bilateral renal cysts. The hepatic flexure wall was thickened. CHIEF COMPLAINT: Vertigo or dizziness. There was sigmoid diverticulosis without diverticulitis. It has been recommended to her that she undergo repair of the ventral hernia. For this reason, cardiology consult is obtained to assess whether she can be cared from the cardiac standpoint. HISTORY OF PRESENT ILLNESS: This is an (XX)-year-old male with a past medical history of coronary artery disease, CABG done a few PAST CARDIAC HISTORY: Bypass surgery. She underwent echocardiography and cardiac years ago, atrial fibrillation, peripheral arterial disease, peripheral neuropathy, recently retired one year ago secondary to leg pain. The catheterization prior to the operation. Echocardiography showed an ejection fraction of 50%. There patient came to the ER for an episode of vertigo while reaching for some books. The patient was able to reach the books, to support was marked left ventricular hypertrophy with septal wall 1.60 cm and posterior wall 1.55 cm. self, but did not have any syncope. No nausea or vomiting. No chest pain. No shortness of breath. Came to ER and had a CT head, Coronary arteriography showed 90% stenosis in the anterior descending artery, situated distally which was within normal limits. The impression was atrophy with old ischemic changes but no acute intracranial findings. No focal just before the apex of the left ventricle. Only mild to moderate narrowing was seen elsewhere in the coronary circulation. weakness, headache, vision changes or speech changes. The patient has had similar episodes since one year. Peripheral neuropathy CORONARY RISK FACTORS: Her father had an irregular heartbeat and her brother had a fatal since one year and not relieved with multiple medications. The patient also complains of weight loss of 25 pounds in the last 6 Report: heart attack. She herself has had high blood pressure for 20 years. She has elevated cholesterol months. NoEchocardiogram colonoscopy done.Sample Recent history of hematochezia but believes it was secondary to proctitis and secondary to decreased and takes Lipitor. She has had diabetes for 20 years. She is not a cigarette smoker. She does little appetite. No nausea, vomiting, no abdominal pain. physical exercise. DATE OF STUDY: MM/DD/YYYY REVIEW OF SYMPTOMS: CARDIOVASCULAR AND RESPIRATORY: She has no chest pain. She sometimes becomes short of breath if she walks too far. No cough. She has occasional swelling of her feet. Occasionally, she gets mildly lightheaded. Has not lost consciousness. She tends to be PROCEDURES PERFORMED: The patient had x-ray, which showed cardiomegaly with atherosclerotic heart disease, pleural DATE OF INTERPRETATION OFa chest STUDY: aware of her heartbeat when she is tired. She has no history of heart murmur or rheumatic fever. thickening and small pleural effusion, a left costophrenic angle which has not changed when compared to prior examination, COPD GASTROINTESTINAL: Recent GI symptoms as noted above, but she does not usually have such pattern. The patient also had a head CT which showed atrophy with old ischemic changes. No acute intracranial findings. Echocardiogram was obtained for assessment of left ventricular problems. She has had no hematemesis. She has no history of ulcer or jaundice. She sometimes has loose stools. No constipation and no blood in the stool. GENITOURINARY: She tends to have function. The patient has been admitted Cardiology with diagnosis of Consultation Transcribed Medical Transcription Sample Reports urinary frequency. She gets up once at night to pass urine. No dysuria, incontinence. She has had syncope. Overall, the study suboptimal due to poor sonic window. CONSULTS OBTAINED: A rehab consult waswas done. previous urinary infections. No stones noted. NEUROLOGIC: She has occasional headaches. No DATE OF CONSULTATION: MM/DD/YYYY REFERRING PHYSICIA N: John Doe , MD seizures. No trouble with vision, hearing, or speech. No limb weakness. MUSCULOSKELETAL: She CONSULTI NG PHYSICIA N: Jane Doe , MD tends to have joint and muscle pains and has a history of gout. HEMATOLOGIC: No anemia, FINDINGS: REA SON FOR CONSULTA TION: Surgical e valuation for coronary arte ry disease . abnormal bleeding, or previous blood transfusion. GYNECOLOGIC: No gynecologic or breast PAST MEDICAL/SURGICAL HISTORY: Positive for atrial fibrillation. patient AVR 6 years ago. Peripheral arterial HISTORYThe OF PRESE NThad ILLNESS: The patient is a (XX)-year-old femdisease ale who with has a k nown history of coronary arte ry problems. disease . She unde rweCABG, nt pre vious A and ste nting procedure s in De ce mbe r and most re ce ntly in August. Since that time , hypertension, neuropathy, atherosclerosis, hemorrhoids, proctitis, and PTC cholecystectomy. 1. peripheral Aortic root appears normal. PAST MEDICAL HISTORY: She has had shoulder and hand injuries and has had carpal tunnel she has been relative ly stable with me dical manageme nt. Howe ve r, in the past se ve ral we e ks, she starte d to notice som e ex e rtional dyspnea with che stispain. For the m ost part, the pain subsides with re st. For this reason, surgery. she She was re has -e valuated been diabetic and has been on insulin. She has chronic renal insufficiency with 2. Left atrium is mildly dilated. No gross intraluminal pathology wi th a cardiac cathe te rization. This demonstrated 3-vesse l coronary arte ry dise ase with a 70% lesion to the right coronary creatinine around 2.2. She has had hypothyroidism. She has had morbid obesity. She has chronic recognized, subtle abnormalities could not excluded. Right arte ry; this wasbe a prox imal lesion. The le ft main had a 70% ste nosis. The circumfle x also had a 99% ste nosis. O ve rall left FAMILY HISTORY: Positivealthough for atherosclerosis, hypertension, autoimmune diseases in the family. obstructive sleep apnea and uses BiPAP. She has had hysterectomy and oophorectomy in the past. ve ntricular function was m ildly re duced with an e jection fraction of about 45%. The le ft ve ntriculogram did note some apical atrium is of normal dimension. Otherwise noted above. hypok ine sis. In vie w of these findings, surgical consultation was re que ste d and the patient was see n and eas valuate d by Dr. MEDICATIONS: Prior to hospital, she was taking glipizide XL 2.5 mg daily, metoprolol 50 mg 3. There is echo dropout of the interatrialDoe. septum. Atrial septal defects PA ST MEDICA L HISTORY: b.i.d., Cipro 250 mg b.i.d., atorvastatin 40 mg daily, Synthroid 75 mcg daily, aspirin 81 mg daily, SOCIAL HISTORY: Alcohol socially. No drugs. could Never not besmoked. excluded. 1. Coronary arte ry disease as described above with pre vious PTC A and stenting procedures. and Lantus 36 units daily. Currently, she is taking Lipitor 40 mg daily, Lantus 10 units at bedtime, 2. Dyslipidem ia. 4. Right and left ventricles are normal in internal dimension. Overall left Synthroid 75 mcg daily, metoprolol 50 mg b.i.d., and Zosyn 2.25 grams q.6h. 3. Hype rte nsion. SOCIAL HISTORY: She does not drink alcohol. ventricular systolic function appears to be4.normal. Eyeball ejection Status post breast lumpe ctom y for cance r with followup radiation the rapy to the che st. ALLERGIES: NO KNOWN DRUG ALLERGIES. PHYSICAL EXAMINATION: A LLERGIES: None. fraction is around 55%. Again, due to poor sonic window, wall motion MEDICA TIONS: Aspirin 81 mg daily, Plavix 75 mg daily, Altace 2.5 m g daily, me toprolol 50 m g b.i.d. and Lipitor 10 mg GENERAL APPEARANCE: She is not currently dyspneic, in no distress. She is alert, oriented, and abnormalities in the distribution of lateralq.h.s. and apical wall could not be pleasant. SOCIA L HISTORY: She quit smok ing approx imately 8 months ago. Prior to that time , she had about a 35- to 40-pack HEENT: Pupils are normal and react normally. No icterus. Mucous membranes well colored. REVIEW OF excluded. SYMPTOMS: Weight loss of 25 pounds within the 6 months, ofalcohol. breath, constipation, bleeding from ye arlast history. She doeshortness s not abuse NECK: lySupple. FA MILY MEDICA HISTORY: Mothe r die d premature ly of breastincance He r fathe r die d premature of gastricNo lymphadenopathy. Jugular venous pressure not elevated. Carotids equal. hemorrhoids, frequency of urination, muscle aches, dizziness andL flow faintness, focal weakness and numbness bothr. legs, knees 5. increased Aortic valve is sclerotic with normal excursion. Color imaging and HEART: The heart rate is 82 per minute and regular and the blood pressure 132/78. The cardiac carcinom a. and feet. Doppler study demonstrates trace aortic regurgitation. REVIEW OF SYMP TO MS: The re is no history of any C VAs, TIAs or se izures. No chronic headache s. Nohas asthma, TB, quality. There is a grade 3/6 ejection systolic murmur heard medial to the impulse a normal hemoptysis or productive cough. The re is no congenital heart abnorm ality or rhe umatic fe ve r apex history. She and athas thenoaortic area, with well heard radiation to the neck vessels. 6. Mitral valve leaflets are also sclerotic with normalSheexcursion. Colorvom iting, constipation, diarrhea, but im mediate ly prior to adm palpitations. note s no nausea, ission, she did de ve lop CHEST: Chest is clear to percussion and auscultation. Normal respiratory effort. some diffuse inal degree discomfort. She says that since then, this has resolved. No diabe tes or thyroid problem . The re is flow imaging and Doppler study demonstrates traceabdom to respirations mild ofand ABDOMEN: Soft and nontender. The presence of a large ventral hernia is noted. PHYSICAL EXAMINATION: VITAL SIGNS: Blood pressure 188/74, pulse 62, 18 saturation of 98% on room air. General no de pre ssion or psychiatric problems. The re is no m usculoske le tal disorde rs or history of gout. The re are no hematologic mitral regurgitation. EXTREMITIES: There is no edema. Posterior tibial pulses were felt bilaterally, but I did not feel the problems or blood dyscrasias. No blee ding tende ncie s. Again, she had a history of breast cance r and underwe nt Appearance: The patient is a pleasant man, comfortable. HEENT: Conjunctivae are normal. PERRLA. EOMI. NECK: No lumpe ctom yPulmonic proce dures valve for this is with followup radiation the rapy. She has been followe d in thedorsalis past 10 years and pedis. 7. Tricuspid is delicate and opens normally. not masses. Trachea is central.valve No thyromegaly. LUNGS: Clear to auscultation and percussion bilaterally. HEART: Irregular mammography shows no e vidence of any re curre nt problems. The re is no re ce nt fe ve rs, malaise , changes in appe or SKIN: No rash ortite significant lesions are noted. clearly seen. No evidence pericardial effusion. rhythm. ABDOMEN: Soft, nontender, andof nondistended. Bowel are positive. GENITOURINARY: Prostate is hypertrophic with change sounds s in we ight. LABORATORY AND DIAGNOSTIC DATA: Electrolytes are normal. BUN and creatinine 18/2.2. PHYSICA L EXA MINA TIO N: He r blood pre ssure is 120/70, pulse is 80. She is in a sinus rhythm on the EKG smooth margin. EXTREMITIES: Upper and lower limbs bilaterally normal. SKIN: Normal. NEUROLOGIC: Cranial nerves are grossly sugar 150. is 7.6, hemoglobin 11.7 with hematocrit 34.9, platelets 187,000. monitor. Re spirations are 18 and unlabore d. Tem pe rature is 98.2 de grees Fahre nhe it. She weBlood ighs 160 pounds, sheWhite is 5 fecount et within normal limits. No nystagmus. DTRs are normal. Good sensation. The awake, and pleasant orientedfemale x3. Mild LFTsdistre were Hemoglobin A1c 7.7. TSH 1.82. Troponin I was normal on three occasions. 4 inches. In ge ne ral, patient this was is analert, elde rly-appearing, who confusion. curre ntly is not in acute ss.normal. Sk in color and CONCLUSIONS: turgor are good. Pupils we re e qual and reactive to light. Conjunctivae clear. Throat is benign. Mucosa was m oist and an enlarged heart with postoperative changes, but no evidence of acute Chest x-ray showed noncyanotic. Ne ck ve ins not distended at 90 degrees. Carotids had 2+ upstrok es bilate rally without bruits.EKG No shows probable left atrial enlargement. Low voltage QRS, probable inferior wall pathology. lym phade nopathy was appre ciate d. C hest had a normal AP diame te r. The lungs we re cle ar in the apices and base s, no 1. Poor quality study. myocardial infarction and anterior wall infarction, age undetermined. LABORATORY DATA AND RADIOLOGICAL RESULTS: WBC 8.6, whe e hemoglobin zing or e gophony 13.4,appre hematocrit ciated. The 39.8, heart platelets had a norm 207,000, al S1, MCV S2. No 91.6, m urmurs, click s or gallops. The abdomen was soft, ASSESSMENT: 2. Eyeballofejection fraction133, is 55%. nontende r, nondiste nded. urea Good nitrogen bowe l sounds No he patosplenom egaly was appre ciate d. No pulsatile masse s we re neutrophil percentage 72.6%. Sodium potassium 4.7, chloride 104. Blood of 18prese andnt. creatinine of 1.1. PT 17.4, INR fe lt. No abdom inal bruits we re heard. He r pulses are 2+ and e qual bilate rally in the uppe r and r ex trem itie s. No 1.lowe Aortic valve replacement with bioprosthetic valve. Residual systolic murmur. 1.6, PTT 33. 3. Trace to mild degree of mitral regurgitation. clubbing is appre ciate d. She is oriented x 3. Demonstrated a good am ount of stre ngth in the uppe r and lowe r 2. Arteriosclerotic heart disease with severe stenosis in anterior descending artery, but this is ex trem ities. Face was sym me trical. She had a norm al gait. 4. Trace aortic regurgitation. situated distally and subtends only a small mass of myocardium. IMPRESSION: This is a (XX)-year-old fe male with significant m ultive sse l coronary arte ry disease . The patient also has a 3.toWell preserved le ft main le sion. She has unde rgone se ve ral PTC A and stenting procedures within the last ye ar year and a half. Atleft thisventricular systolic function. The EKG appearance of previous myocardial The patient had a chest x-ray, which showed cardiomegaly point, with atherosclerotic disease, pleural thickening andfuture small, surgical pleural m yocardial reinfarction is probably serious, indicating multiple other medical problems as listed above in orde r to re duceheart the risk of any possible ischem ia in the vascularization is

DATE OF ADMISSION: MM/DD/YYYY

Ongestructureerde data zijn rommelig maar vol van relevante medische gegevens

DATE OF DISCHARGE: MM/DD/YYYY

Medicatie, aandoeningen, symptomen, labwaarden, sociale historie, familie geschiedenis, …

2014 International Business Machines Corporation re commende d. 8 effusion,©a left costophrenic angle which has not changed when compared to prior examination, COPD pattern. The patient also had a and also documented in the chart. PLA N: W e will plan to procee d with surgical m yocardial re vascularization. The risks and be ne fits of this procedure we re head CT, which showed atrophy with old ischemic changes.exNo acute findings. RECOMMENDATIONS: It appears that she does not wish to proceed with the plained to intracranial the patient. All questions pe rtaining to this proce dure we re answe re d. surgery at this time, and if such surgery is not

Zorg op het lijf geschreven - realiteit vereist betere toegang tot en analyse van relevante patiëntinformatie en klinische kennis Meer kunst dan wetenschap Meer wetenschap dan kunst

Goed

Gepersonaliseerd Zorginformatie verdubbelt elke 5 jaar

(Gebaseerd op mensen zoals ik)

Toegang tot relevante patiëntinformatie

Wa

e ard

Voorspellend en Evidence-based

80% van de artsen hebben hoogstens 5 Intuïtief en volgens klinische consensus uur/maand om hun vakliteratuur bij te houden (Gebaseerd op patiëntencohorts)

(Op basis van partiële toegang tot beschikbare patiëntinformatie en klinische kennis)

Proefondervindelijk (Gebaseerd op expertise en ervaring) Matig

1 op de 5 diagnosen is niet Toegang tot klinische kennis accuraat of onvolledig Matig Goed (e.g. Diagnostische hulpmiddelen, kennis van de oorzaken van ziekten, empirisch bewijs of vergelijkende effectiviteit)

Bron: IBM Global Business Services and IBM Institute for Business Value


9

Waar is de holistische kijk op de patiënt?

Symptomen – Koorts, duizeligheid, buikpijn, rugpijn, hoest, ...

Familiehistorie – Diabetes, borstanker, hartproblemen, erfelijke ziekten, ...

Patiënthistorie – Hoge bloeddruk, verhoogd cholesterolgehalte, verminderde schildklierwerking, chronische blaasontsteking, roken, alleenwonend, ...

Klinische historie – Bloeddruk, hartslag, temperatuur, pijnscores, labuitslagen, ...

Medicatiehistorie – Pravastatine, betablokkers, lasix, aspirine, chemotherapie, antiemetica, ... © 2014 International Business Machines Corporation

10

Ontworpen voor de uitdagingen in taal en betekenis food would “get stuck” when she was swallowing

sudden onset of chills

…can cause food to move Abnormal QRS Complex slowly in the esophagus. Delta-Wave

swallowing difficulty…

Terminology

PR Interveral

Domain

…food gets held-up…

coryza

cold

chills

Terminology

productive cough after nonproductive cough

Fever

Magnitude High Temperature

Chronology

Normal QRS Pattern

Causation

Temperature

Fever after acute symptoms Chronology subside…

Negation productive cough

nonproductive cough

Chronology

Abdomen Pain

Terminology Urination Pain

Causation

between the upper abdomen and the back

pneumaturia bubbles in the urine

Dysuria

Flank Pain

Location Lower Back Pain

Kidney Pain

Abdomen Pain exacerbated by exercise

11


Gebruikt de New England Journal of Medicine voor annotaties van medische concepten Aandoening Symptomen

Modificatoren Medicatie


12

Big Data: meer dan de traditionele datatypen Transactionele - en applicatiedata Social Media data Enterprise Content Wetenschappelijke artikelen Medische beelden Klinische rapporten

13

© 2014 International Business Machines Corporation 13

De Megatrends Cloud Mobile

Social

Analytics © 2014 International Business Machines Corporation

14

Big Data – meer dan de volume-dimensie!

Volume

Petabytes Exabytes Zetabytes Iotabytes Brontobytes Geobytes

Velocity

Variety

Streaming data, milliseconden tot seconden tijd om te antwoorden

Gestructureerd, ongestructureerd tekst, multimedia

Veracity

Onzekerheid, inconsistensie, ambiguiteiten

15


Watson luidt een nieuw computertijdperk in

Tabellerende Systemen

Programmeerbare Systemen

Cognitieve Systemen

1900

1950

2011

Systemen welke: In en op natuurlijke taal reageren Oceanen van gestructureerde en ongestructureerde gegevens verwerken Van elke interactie leren © 2014 International Business Machines Corporation

16

Voor een computer is taal pas echt lastig!

Schaken (Deep Blue, 1997) – Een eindige welgedefinieerde afgebakende zoekruimte – Beperkt aantal zetten en situaties – Gebaseerd op expliciete, eenduidige wiskundige regels

Natuurlijke taal – Meerduidig, context afhankelijk en impliciet – Gebaseerd op menselijk cognitief vermogen – Ogenschijnlijk een oneindig aantal manieren om hetzelfde uit te drukken 17


Commercialisatie van Watson vereist een aantal substantiële stappen Watson aan het spelen 1 Gebruiker 1 Vraag Max. input – 2 zinnen Stateless Focus op beste antwoord 1 week hertrainen

Watson aan het werk 1000en Gebruikers tegelijkertijd Samengestelde vragen Uitgebreide tekst input Statefull Nadruk op beste suggesties Dynamische invoer van content

Bewijs niet aanwezig

Volledige traceerbaarheid

Alleen tekstinvoer

Tekst, tabellen, plaatjes, …

Q&A omgeving Elementaire beveiliging © 2014 International Business Machines Corporation

Q&A + Case Mgmt. omgeving Privacy & security (e.g. HIPAA, NEN 75xx, …)

18

De adviseur van de arts Complexe diagnosen Betere en beter afgestemde medicatie Voorkomen van medische missers Evidence Based Medicine (EBD) Natuurlijke taal interface – NLP Opstellen van hypothesen Zelf-lerend en adaptief

19


Hoe Watson werkt: DeepQA architectuur Learned Models help combine and weigh the Evidence Evidence Sources Answer Sources

Question

Primary Search

Candidate Answer Generation 100’s Possible Answers

Balance Models & Combine

Models

Models

Models

Models

Models

Deep Evidence Answer Evidence Retrieval Scoring Scoring 100,000’s Scores from many Deep Analysis 1000’s of Algorithms Pieces of Evidence

100’s sources Question/Topi Multiple c Analysis Interpretations

Question Decomposition

Hypothesis Generation

Hypothesis and Evidence Scoring

Synthesis

Final Confidence Merging & Ranking

of a question



Hypothesis and Evidence Scoring

Responses with Confidence

20

De componenten van Watson Decomposition

Hypothesis & Evidence Scoring


Synthesis

Final Confidence Merging & Ranking

Workload Optimized

Question & Topic Analysis

21


Het Watson Platform Specialized Watson Advisor Solution T in to loin AnswersOnra g Da gie s ta e struc tur Un Abbreviati onsd

idenc Ev Lex s e ic on

T in to loin AnswersOnra g Da gie s e ta struc tur Abbreviati Un onsd

St ruc

tur e

d

idenc Ev Lex s e ic on

Data

Algorithms / NLP

Domain Accelerators

Tooling Extensions

Data

User Experience / Watson Enabled App

(Healthcare, Financial Services, Contact Center, etc.)

d

tur e

St ruc

Integration Services

Algorithms / NLP

Methodology

Tooling Extensions

User Experience Patterns

General Watson Platform Core St d r uctu re

Onto Tra inin c loD g Eviden gie Answers s a ta Lex ture AbbreviData ico ation s Unstruc n s d

Infrastructure, Ops, Lifecycle & System Management

Algorithms / NLP

Tooling – Accuracy & Solution Construction

User Experience Patterns / Watson Enabled App Toolkit

Methodology

Content Creation & Management

Runtime

Extensability / Programmability

Ready

Full Lifecycle © 2014 International Business Machines Corporation

Build

Teach

Run

22

Solution

Watson Referentie Architectuur Watson for Industry Watson for Healthcare

Watson for Financial Services

Watson for Contact Center

User Experience / User Interface

Product

NOW

Watson Decision Advisor

NOW

NOW

Future

Watson Engagement Advisor

Watson Diagnose & Treatment Advisor

Watson xxx Advisor

DISCOVER Services

DECISION Services

100111001 10010010010 1000101100101 10001010010 00110101

Data

NLP & Machine Learning

Foundation

Technology

ASK Services

Watson future Advisor

Content

Analytics

Tooling

Cloud

Methods

Workload Optimized Systems

Mobile

Algorithms

APIs

23


2013 Annonceringen van Watson Eerste twee commerciële producten – 8 februari 2013

Watson Ecosystem – 14 november 2013 Watson Ecosystem Developer Cloud, Content Store, APIs T.b.v. innovatie en het cultiveren van nieuwe mogelijkheden

Eerste twee zorgproducten geannonceerd i.s.m. WellPoint en MSK, Watson Decision Advisor, Research & Insight Advisor Januari 2013

April

Juli

Oktober

December 2013

Watson Engagement Advisor – 21 mei 2013

Watson Healthcare Update – 18 oktober 2013

Watson Engagement Advisor geannonceerd Cross-industry en crossgeography

MD Anderson lanceert de Watson Diagnosis & Treatment Advisor Watson wordt gebruikt om een brug te slaan tussen klinische praktijk en medisch onderzoek


24

Watson Engagement Advisor Helpt organisaties de diensten naar hun klanten te verbeteren. Betrekt klanten via een dialoog bij zoeken naar informatie en het nemen van beslissingen.

Call centers Banken Verzekeraars … 25


25

IBM Oncology Diagnosis & Treatment Advisor Laat zien hoe Watson een oncoloog kan assisteren bij: Correlatie van verspreide data, zoals patiëntendossiers, professionele samenvattingen, testresultaten, pathologie rapporten, etc. Identificatie van ontbrekende gegevens en aanbeveling om aanvullende onderzoeken te doen. Suggereren van patiëntgepersonaliseerde, gewogen, evidence-based behandelopties


26


27


28


29


30

Creating a Corpus of Knowledge for Cancer Care Ingestion of NCCN guidelines for breast cancer and lung cancer: – –

Roughly 500,000 unique combinations of breast cancer patient attributes. Roughly 50,000 unique combinations of lung cancer patient attributes.

Over 600,000 pieces of evidence ingested, from 42 different publications/publishers, including: –

The Breast Journal, National Comprehensive Cancer Network (Clinical Practice Guidelines, Drug and Biologics compendium, et al.), American Journal Of Hematology, Annals Of Neurology, CA: A Cancer Journal For Clinicians, Cancer Journal, Cochrane, EBSCO, Hematological Oncology, Hepatology, International Journal Of Cancer, Journal Of Gene Medicine, Journal of Clinical Oncology, Journal of Oncology Practice, Massachusetts Medical Society Journal Watch, Massachusetts Medical Society New England Journal Of Medicine, Merck, Nephrology, UptoDate, Clinical Lung Cancer, Current Problems in Cancer, Cancer Treatment Reviews, Elsevier's Monographs in Cancer (multiple), Clinical Breast Cancer, European Journal of Cancer, Lung Cancer (the journal).


31

IBM Confidential

Nieuw in 2014 IBM Watson Business Unit - HQ in New York Commercialisatie van cognitieve producten en diensten – Investering van $1 miljard over een aantal jaren – 2000 professionals, – $100 miljoen equity investment

Nieuwe Watson producten


2880 cores Single user system 2-3 sentences input 5+ days to retrain Wikipedia, general corpus

Single Power 750, 240% faster 1000s of users 20 pages of input < Day to ingest and train Medical corpus

#users

flexibility

corpus

speed

size

#users

flexibility

corpus

speed

Size

#users

flexibi

corp

sp

s

– Marketing research – Visualisatie

Watson cloud serv ice Millions of users Dialog chaining Few hours Broad industry corpus32

Watson Hoofdkantoor 51 Astor Place, Manhattan (Silicon Alley)

Incubator voor innovatie Client Solutions Center Interactieve hands-on Design Lab Workshops en Seminars


33

Watson Analytics Maakt via natuurlijke taal data selecties, semantische en mathematische analyse op de achtergrond Om nieuwe patronen snel te herkennen via een eenvoudig NLP en visueel interface betere beslissingen snel te nemen Focus op Sales & Marketing Cloud-based model 34


34

Watson Explorer Visualisatie exploratie software om mensen te helpen met het begrijpen van wat er in hun gegevens schuil gaat Vinden, ophalen en afleveren van content ongeacht het formaat of waar het zich bevindt Gestructureerde data, ongestructureerde content, e.d..

35


35

Hoe met Watson aan de slag te gaan?

Workshop – Watson of de “route naar Watson”?

IBM Global Business Services (GBS) – Watson Assessment Workshop

IBM Software Groep IBM Center for Advanced Studies (CAS) – Onderzoek i.s.m. VU Infomatica, Amsterdam


36

Technische Routekaart

Ontsluiting databronnen (taal!), content ingestie (e.g. Patiëntregisters) Consolidatie, opschoning en verrijking van de data Visualisatie middelen Content sharing Annotations en Rules Trainen van Watson Connections Predictive tooling

37


Wanneer is het nu Watson? 1/2

Cognitief Stochastische optimalistaie

Business Value

Natuurlijke taal interactie met evidence based, confidence weighted antwoorden, leren van nieuwe inzichten

Vertel me wat ik het beste kan doen

Cognitief

Wat is de beste optie gegeven de variabiliteit? Voorschrijvend

Optimalistaie

Hoe kunnen we de beste uitkomst krijgen?

Modelleren

Wat zal er gebeuren als …?

Simulatie

Wat kan er gebeuren?

Voorspellen

Wat als deze trends aanhouden?

Alarmsignalen

Welke acties zijn nodig?

Analyse

Wat is het probleem precies?

Ontdekking

Hoe veel, hoe vaak, waar?

Rapportage

Wat gebeurde er?

Voorspellend


Beschrijvend

38

Wanneer is het nu Watson? 2/2 Het is Big Data wanneer…

Het is BI (Cognos) wanneer…

Het is SPSS wanneer…

• Dealing with large complex and dynamic data sets

• Need is for reporting, dashboards & scorecards -“Business Intelligence”

• Desire is to leverage predictive outcomes

• Looking to coordinate structured and unstructured data • Interest is in streaming data • Looking to find rare events, common patterns, and outliers • General focus is on data architecture, extraction, transformation, movement, storage, integration, and governance

• Looking to spot and analyze trends, patterns and anomalies • In-memory reporting and analysis is required • Need is to perform enduser query and analysis • Requirements are for planning, budgeting and forecasting resources

• Data mining is required for deep insights and pattern recognition • Interest is in using statistics including logistic regression. • Looking to create predictive scores around key attributes • Automating decision management processes leveraging predictive analytics and business rules for optimization.

Het is Watson wanneer... • Looking to perform question and answer iterations with confidence based responses • Need to work with unstructured data, leveraging natural language processing to deal with complexity of human speech. • Hypothesis generation and validation is critical to the process • Evidence based learning is integral to the solution delivered

39


De 5 kernfunctionaliteiten van Watson Analyseert grote volumes van gestructureerde en ongestructureerde data

Combineert grote hoeveelheden ongestructureerde met gestructureerde gegevens om samen te worden geanalyseerd

Interpesteert en begrijpt vragen gesteld in een natuurlijke taal

Begrijpt dubbelzinnige en onduidelijke vragen met behulp van geavanceerde natuurlijke taal algoritmen

Genereert en evalueert hypothesen en kwantificeert het vertrouwen in de antwoorden

Identificeert veel antwoorden op vragen met bewijs om de rationaal te verklaren

Ondersteunt een iteratieve dialoog om de antwoorden te verfijnen

Stelt iteratieve en interactieve vragen ter verfijnening verbetering van de resultaten

Past zich aan en leert om in het vervolg met nog betere resultaten te komen

Leert van aanvullend bewijs, aanvullende vragen en fouten om in de tijd de nauwkeurigheid te verbeteren


40

IBM Content and Predictive Analytics for Healthcare Extract medical facts and relationships from multiple clinical and operational information sources Analyze and Visualize past, present and future scenarios to create an evidence based corpus of information Enable clinicians, executives and knowledge workers to Interact with information and derive insight in new ways

Integrate and leverage other systems to turn insight to action Raw Information

Unstructured Data (Nurses notes, discharge notes, etc.)

Structured Data (Billing data, EMR, etc.)

Analyzed and Visualized Information

IBM Content and Predictive Analytics Content Analytics

Predictive Analytics

• Natural Language Processing • Medical Fact and Relationship Extraction (Annotation) • Trend, Pattern, Anomaly, Deviation Analysis

• Predictive Scoring and Probability Analysis

Dynamic Multimode Interaction

Search and Visually Explore (Mine) Monitor, Dashboard and Report Question and Answer*

Health Integration Framework

Data Warehouse and Model

Custom Solutions

Master Data Management Advanced Case Management

* Future optional capability

Partners (HLI)

Specialized Research

Business Analytics

41


Slimmere gezondheidszorg d.m.v. analytics

Integratie van virale genomics data met klinische gegevens om de reactie op een antiHIV therapie te voorspellen.

Real-time analytics op streaming data van medische instrumenten ter monitoring van te vroeg geboren baby's in het Toronto Sick Children's Hospital.

Verbetering van kennis over en plannen van heropnames rondom Congestief Hartfalen (CHF) via het correleren van ongestructureerde en gestructureerde data. © 2014 International Business Machines Corporation

Koppeling van o.a. DNA-expressies en klinische data voor een beter begrip van en sneller onderzoek naar inflammatoire darmziekten.

Verbetering van kennis en diagnostiek en vroegtijdige herkenning t.b.v. zeldzame ziekten.

Creëren van een uitgebreide corpus van gegevens over Epilepsie, waaruit via predictive analytics betere behandelingen en medicatievoorschriften worden gedistilleerd. 42

Seton - wat werd er gevonden? De gegevens waarvan men dacht dat ze nuttig waren, waren dat niet! •

113 kandidaat-predictoren uit gestructureerde en ongestructureerde data bronnen

•

Gestructureerde gegevens waren minder betrouwbaar

Nieuwe onverwacht indicatoren kwamen naar boven drijven 18 accurate indicatoren of predictors

•

Predictor Analysis

% Encounters Structured Data

% Encounters Unstructured Data

Ejection Fraction (LVEF)

2%

74%

Smoking Indicator

35% (65% Accurate)

81% (95% Accurate)

Living Arrangements

<1%

73% (100% Accurate)

Drug and Alcohol Abuse

16%

81%

Assisted Living

0%

13%

43


Seton - top 18 indicatoren

Ranking of Strength of Model Variable

LVEF en Roken zijn significante indicatoren van CHF maar niet van heropnamen. Assisted Living en Overmatig Drugs en Alcohol Gebruik bleken de belngrijkste predictors (alleen aangetroffen in ongestructureerde data) Vele predictoren werden in historische en observationele aantekeningen gevonden 18. Jugular Venous Distention Indicator 17. Paid by Medicaid Indicator 16. Immunity Disorder Disease Indicator 15. Cardiac Rehab Admit Diagnosis with CHF Indicator

18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

14. Lack of Emotion Support Indicator 13. Self COPD Moderate Lim it Health History Indicator 12. With Genitourinary System and Endocrine Disorders 11. Heart Failure History 10. High BNP Indicator

0

1

2

3

4

5

6

Projected Odds Ra o 18 9

17 8

16 7

15 6

14 5

13 4


12 3

11 2

10 1

9. Low Hemoglobin Indicator 8. Low Sodium Level Indicator 7. Assisted Living (from ICA Extract) 6. High Cholesterol History 5. Presence of Blood Diseases in Diagnosis History 4. High Blood Pressure Health History 3. Self Alcohol / Drug Use Indicator (Cerner + ICA) 2. Heart Attack History 1. Heart Disease History

44

Regio Montpellier: reumatoïde polyartritis

Chronische ziekte – aantasting van het autoimmuun systeem, ontsteking van het synoviaal vlies Invaliderende misvorming van gewrichten – vnl. handen en voeten Zeer belangrijk dat de behandeling aangepast wordt aan de patiënt, want polyartritis uit zich niet bij iedereen op dezelfde manier

De ziekte kent hoge en lage fasen Erosieve en niet-erosieve vormen Een goede vroege behandeling kan leiden tot een lange lage fase

IBM La Gaude Global Healthcare Center of Excellence gebruikte ICPA Annotaties van Franse medische termen Ontwikkelen van een prototype

Het vinden van de risico predictoren bij het eerste bezoek van de patiënt 45


Architectuur Data Data Warehouse Warehouse Longitudinal Longitudinal patient patient record record

Extraction, integration

Share knowledge

IBM Cognos BI - Sharing reports - Operational dashboards

Medical & operational optimisation actions Structured & Unstructured datasources

Metadata

IBM Content Analytics - Natural language processing jobs - Data mining Identification of facts, correlations © 2014 International Business Machines Corporation

Predictive data

IBM SPSS Modeler - Building a predictive model - Scoring jobs Scoring based on historical data 46

Watson pilot project in het UMCG Onderzoek naar hoofd-hals kanker

Zoeken in gestructureerde en ongestructureerde data naar correlaties tussen hoofd-hals kanker en andere aandoeningen (comorbiditeit) Opstellen van een business case voor gebruik in de praktijk van dit soort technieken Content invoer, annotaties en rules – Cohort van 600 geanonimiseerde patiënten (19942010) – Ontslagbrieven, verwijsbrieven, medicatiedossiers, etc. – DHD medische termen – Behandelingen, diagnose methodieken, protocollen, e.d.

Technologie: IBM Content en Text Analytics


47

Met dank voor uw aandacht


48


49

The Singularity is near When humans transcend technology

De Wet van de Versnellende Meeropbrengst voorspelt een exponentiële toename in technologie – computers, genetica, nanotechnologie, robotica, kunstmatige intelligentie.

Dit leidt tot een technologische singulariteit in 2045 – Versnelling gaat harder dan de mens kan bevatten. – De mens zijn lichaam en geest uitbreiden met genetische veranderingen, nanotechnologie, en AI.


50

Watson Assessment BVA is a rapid workshop based diagnostic Objectives: Conduct kickoff meeting with project sponsor and key stakeholders Identify business issues to be explored during the Watson Assessment Ensure common understanding of the project including expectations, work team roles and responsibilities Finalize Watson workshop agenda, participants, and logistics

Mobilize Understand business priorities and Watson interest areas

Outcomes: Watson workshop scope, agenda, participant list, and expected results

Timeline: 1 week before workshop

Objectives: Conduct Watson overv iew Understand client context for Watson and categorize opportunities Explore analytics usage in areas such as Customer, Marketing, and Sales; Risk, Finance, & Fraud; & Information Management Foundation Assess business issues and collaborate with BAO subject matter experts

Explore Conduct Watson workshop and BAO envisioning session

Outcomes: Watson workshop findings

Timeline: 1-2 days

Act

Objectives: Confirm Watson workshop and BAO envisioning findings with sponsor Prepare recommendations and action plan Conduct executive rev iew with key stakeholders

Prepare Watson assessment and BAO action plan

Outcomes: Watson Assessment Prioritized set of initiativ es and actionable next steps

Timeline: 1-2 days after workshop

51

51


Step 1.0 Mobilize Tasks Step 1.0 Mobilize

Identify key client focus areas, confirm project expectations, and develop Assessment & Envisioning Workshop agenda, list of participants, and logistics. Conduct interviews with key stakeholders, establish list of pain points, and possible analytics use cases to be included in the Assessment Workshop.

1.1

Identify client sponsor and conduct initial meeting to understand Watson interest areas and 2-3 topics that will be in-scope for the project. Identify all stakeholders and confirm expected project outcomes/deliverables for each step

1.2

Gather all relevant background materials (information on BAO related projects, key strategic initiatives, etc.) and setup stakeholder interviews

1.3

Conduct interviews with key stakeholders to identify pain points, potential analytics use cases, and critical information sources (both current and future) related to the 2-3 identified topics

1.4

Review Assessment & Envisioning Workshop agenda, participants, and expected outcomes with key stakeholders

1.5

Plan and finalize workshop agenda, logistics, and gain commitment on attendance from key stakeholders

Deliverable Assessment & Envisioning Workshop Plan including: 1.

Definition of 2-3 subject areas that will be in-scope for the project

2.

Initial list of pain points, possible analytics use cases, and data integration points for each of the subject areas

3.

List of participants and meeting logistics

Key Resources: 1.

BAO SME / Project Lead

2.

Industry SME


52

Step 2.0 Explore Tasks

Step 2.0 Explore

2.1

Conduct Assessment & Envisioning Workshop

2.2

Document Assessment & Envisioning Workshop results

• Assessment & Envisioning Workshop is a 1-2 day session that includes client and IBM resources. The agenda and subject areas covered will be determined during mobilization. • The session can be conducted at the client location or at an IBM solution center (depending upon scheduling and availability).

Conduct Assessment & Envisioning Workshop with key stakeholders, gain insight on pain points, analytics use cases, and data integration points for selected topic areas. Document Assessment & Envisioning Workshop findings and agreed upon information management foundation and analytics priorities.

Deliverable Documented Assessment & Envisioning Workshop results including: 1.

Agreed upon set of a Watson and analytics priorities

2.

Workshop notes and key findings

3.

List of follow-up actions including task owners and expected timing

Key Resources: 1.

BAO SME/Project Lead

2.

Industry SME

3.

BAO SMEs – (vary based on client subject areas)

4.

IBM Research Advocate (if appropriate)

53


Step 2.0 Explore: Assessment & Envisioning Workshop Agenda Sample Workshop Topics

Sample Workshop Agenda

Watson Overview

Working session featuring 2-3 subject areas identified by the client

08:00am – 08:30am

Welcome & Introductions

08:30am – 10:00am

Watson Overview

For each subject area, agenda is as follows:

10:00am – 10:15am

Break

–

Confirm pain points and client priorities

10:15am – 11:45am

Client Subject Area 1 *

–

Review analytics use cases

11:45am – 12:30pm

Lunch

–

Gain insight on data sources, corpus of data, and integration points

12:30pm – 02:00pm


02:00pm – 02:15pm

Break

–

Next steps

Action planning & Wrap-up


02:15pm – 03:45pm


03:45pm – 04:30pm

Action Planning / Wrap-up

54

Step 3.0 Act Tasks Step 3.0 Act

Develop BAO action plan that includes immediate projects and a list of strategic initiatives to address the requisite technology, data, and resource capabilities.

3.1

Review Assessment & Envisioning Workshop findings and conduct follow-up interviews

3.2

Conduct Technology / Architecture assessment based on workshop findings

3.3

Identify initial Information Management Foundation (IMF) capabilities and recommend IMF requirements

3.4

Create BAO Action Plan which includes a prioritized list of potential projects for each subject area included in the workshop

3.5

Review BAO Action plan with project sponsor and key stakeholders

3.6

Gain commitment to proceed on selected projects

Deliverable BAO Action Plan - High level program plan with a set of: 1.

Immediate actionable projects

2.

Strategic program definition supported by selected Business Value Accelerators

3.

Recommended set of capabilities (technology, skills, etc,) to support analytics & IMF

4.

Targeted list of initiatives based on items identified during the Deep Insight project

Key Resources: 1. 2. 3.


BAO SME/Project Lead Industry SME BAO SME Reviewers

55

Watson in Healthcare & Lifesciences Op weg naar Cognitive Computing

Recommend Documents