AI Doctor based on Department of Computer Science & Engineering, Evolutionary Knowledge Machine (EKM) Kyung Hee University KOREA February 19 th , 2019 Prof. Sungyoung Lee http://uclab.khu.ac.kr
2 Contents / I. Evolutionary Knowledge Machine (EKM) II. AI Doctor Platform (IMP) III. Case Studies (Silo)
Evolutionary Knowledge Machine
4 What is our Research Goal? / AI Doctor
5 Challenges of AI Doctor / Evolutionary Knowledge Big Knowledge
6 What Problems of Big Knowledge? / Knowledge Representation
7 How to obtain Qualified Medical Knowledge? / Data Driven Knowledge Acquisition Expert Driven Knowledge Acquisition
8 Research Areas of AI / source: http://www.legaltechnology.com/latest-news/artificial-intelligence-in-law-the-state-of-play-in-2015 /
9 / White Box and Black Box Model (Supervised Learning for classification) Black Box Models White Box Models Description Description Machine learning algorithms which produce decision Machine learning algorithms which produce decision models in models in such a form that are interpretable for the the form of a set of mathematical functions that are non- domain experts interpretable for the domain experts Decision Tree Decision Rules Neural Networks Graphical Models A set of Neurons are stacked in a Each attribute is a node, A model is determined A probabilistic model for multi-layer form to generate a non- important attributes are piece-wise by a set of reflecting dependency between linear mapping from input to placed higher in the 'rules' that each cover part a set of random variables decision tree output of the problem at hand H(1) H(2) A B Output Random Variables and Rule 1 Root node Input their probabilistic Rule 2 Ordered Layer graph Rule 3 Rules List C D Rule 4
/ Abstract of EKM 10 Structured Data Intervention of Domain Knowledge 3 Expert 2 4 1 Big Data Engineering Support Tool Adaptive Recommendations Knowledge Extraction Alerts Reminders Inference Engine Verification ? Evolutionary Machine Learning Knowledge base
11 Hybrid knowledge Acquisition Model / Research Concept & Scope Input Output CNN Hybrid Knowledge Acquisition Classification Label concatenation J48, Data Driven Rule Image Data Black Box Random Forest, DT Case Knowledge White Box Generation Consolidation & (RDR Transformation) Inferencing Feature Selector Feature Vector (UFS, FCBF, PSO-FS) EMR Data (Structured) RDR Rules Knowledge Authoring Tool Expert Driven DT Rule *RDR: Ripple Down Rules DT Editor Expert *DT: Decision Tree Heuristics RDR Rule RDR Editor White Box Legacy Knowledge Base
12 Features of EKM Knowledge Base /
AI Doctor Platform (IMP)
14 AI Doctor Platform / AI Doctor (Intelligent Medical Expert System) Evolutionary Knowledge Base -> Big Knowledge Management Novel Knowledge Model -> Incremental Knowledge Model High Quality Knowledge -> (Data Driven + Expert Driven) Engineering Tool Support -> Construction, Maintenance, V&V Intelligent Medical Services (Silo)
/ AI-Clinical Decision Support System (CDSS) 15
/ Requirements of AI-CDSS Knowledge Base 16 Medical Knowledge is used in the wood-grain, being sure to keep the freshness, adaptability, and always will o be, to be able to be present in most commercial reliability medical expert system knowledge base does not have these characteristics Knowledge base requirements Detail CASE Final requirements Rule DB built without sufficient medical There should be no Rule (knowledge generation) is flawed Integrity knowledge — Rule should reflect the complete medical knowledge Nonsense — the Rule is an actual medical environment there should be no shortage of ships to Depending on the Medical staff equipment the hospital has, Rule must be customized according to hospital environment - It should the test method / directly Knowledge Adaptability be customizable according to the situation of available resources treatment method / be able to create (medical equipment, inspection equipment, etc.) by medical operation method varies. and maintain environment An update to a new Easy to update new knowledge Freshness rule - Each time a new treatment, prescription, or surgical procedure are Easy to handle derived, the rule should be updated on these matters. It should be based on Must have sufficient credibility Evidence papers, clinical trial data Reliability - Knowledge to be used should be sufficiently reliable by certified from pharmaceutical Based papers, clinical trial data, or EMR inference data . companies, and EMR inference data
/ Existing CDSS and their Features 17 Standard Hybrid Features Big Data Big Data Rule-based Adaptive UI Compliant HIS Integration Knowledge Systems Support Analytics Recommendation Support Storage Acquisition McKesson: InterQual X X O O X O X Clinical Decision Support White Box AllScripts:Knowledge- X X O X X O X Based Medication Administration (KBMA) X X O X O O X Visual DX O X O O X X X Kinesia 360 Black Box Reed Group: MDGuideline X X X O O O O Intelligent DSS Medaware System Personalized X X X O O O O CDSS Gray Box x X O O O O O IBM Watson Incremental knowledge Tool Support for Knowledge Required Features Dialogue Support Personalization Support Maintenance Acquisition https://glneurotech.com/kinesia/products/kinesia-360/ https://www.changehealthcare.com/solutions/interqual https://www.mdguidelines.com/ https://www.allscripts.com/news-insights/blog/blog/2016/01/safer-medications-with-closed-loop-delivery https://www.ibm.com/watson/ https://www.visualdx.com
AI Doctor (Intelligent Medical Platform) Environments / 18 Analytics Tool Blood pressure Knowledge Engineering device Smart Watch Patient 8 Medical Services Silo Tool Knowledge base Thyroid Cancer Physician Silo Sleep Cardiovascular Monitoring Head & Neck Device Physician Silo Cancer Silo Epilepsy Glucose Meter Diabetes Intelligent Silo Silo Medical Services Intelligent Medical Unstructured Text Platform Heterogeneous Input Data Medical PACS Lifelog ENT Lung Cancer Big data Storage (Ear, Nose, Throat) Silo Silo UI/UX Authoring Patient Profile UX Expert Tool Physician Patient Healthlog Public Health Silos Evidence Support Tool EMR/EHR
AI Doctor Platform (IMP) / 19 E K M
SaaS implementation for IMP / 20
21 Uniqueness: Adaptive Services / Adaptive R Recommendation Adaptive E Education Adaptive Q &A Q&A Evolutionary Adaptive (Personalized) Services Knowledge Base
22 Uniqueness: Platform + Engineering Tool / Incremental Learning-based Overall UX quantification over time validation and verification Adaptive UI based on UX Intelli-sense support 01 Knowledge Authoring UI/UX Tool Authoring Tool Physician 04 UX Expert 02 Real time monitoring Evidence support form PubMed Health-log visualization Quality assessment retrieved Data documents Analytics Evidence Tool Support Patient 03 Tool Physician Physician Easy to commercialize by providing development environment
IBM Watson Oncology vs. EKM / 23 Characteristics/Features IBM Watson Oncology Evolutionary Knowledge Machine Based on pre-curated Knowledge Acquisition Process Incremental Learning Model annotations Primarily focused on medical Knowledge Generation from multi-modal Images and textual data (clinical Knowledge Modeling Approach data sources (EMR, clinical notes, medical notes, doctor notes, patient case images, expert heuristics) report) Expert-friendly knowledge authoring Medical Expert Assistance in Limited support for direct environment for incorporating expert Knowledge Creation knowledge incorporation heuristics Knowledge Maintenance Complex and time consuming Seamless knowledge maintenance (RDR) Capabilities Evidence backed Treatment Supported Supported Supports knowledge sharaeability by All learned knowledge is tightly Knowledge Shareability converting knowledgebase into medical coupled with the system logic module https://www.ibm.com/us-en/marketplace/ibm-watson-for-oncology
Case Studies
25 IMP Service (Silo) / Expert A silo provides intelligent medical Heuristics services for specific diseases Clinical Diagnosis Guideline Mind Map Silo Construction Process Mind map creation Silos Treatment Decision Tree Transformation Plain Rule Creation Implementation Follow-up Knowledge Execution and Evaluation Published Research EMR/EHR
26 Knowledge Creation and Diagnosis Recommendation / 5 Knowledge Physician Engineer 4 1 2 3 지 Mind Map Decision Tree Rules Knowledge Base Recommendation Production Rules
Case #1: Cardio- Knowledge Acquisition (Decision Tree) / 27 Decision Tree Mind Map Enterprise Architect Knowledge Engineer Medical Experts
Case #1: Cardio - Knowledge Acquisition (Production Rule) / 28 Production Rules Decision Tree Intelligent Knowledge Authoring Tool of IMP (I-KAT) Medical Experts Total Rules: 1,309 Total Patient Data : 300 Initial Accuracy : 90%
Case #1: Implementation of Cardio - Dashboard / 29 Dashboard: Shows all the patient data from EMR and EHR systems Add New Patient Search Patient Update Existing Patient Patient abstract information Delete Existing Patient
Recommend
More recommend
