HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Intelligent Systems in Healthcare
Janusz Wojtusiak, PhD George Mason University
Spring 2014
HAP 752 Advanced Health Informa6on Systems
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Any sufficiently advanced technology is indis6nguishable from magic.
-‐ Arthur C. Clarke
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
What is Intelligence?
• IQ scores? • GRE scores? • Driving ability? • General knowledge? • Math skills? • Playing chess?
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
What is Intelligence? • Abstract thinking • Reasoning • Memorizing • Learning • Understanding • Planning • Problem-‐solving • Communica6ng • Self-‐awareness • Emo6onal knowledge
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
What is Ar6ficial Intelligence?
• AI concerns intelligent behaviors in machines – OYen machines in AI are called agents
• Intersects mul6ple disciplines – Computer science, philosophy, engineering, psychology, cogni6ve science, economics, mathema6cs, neuroscience, linguis6cs, ….
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Intelligent Systems
• Intelligent systems use AI methods to perform be[er than simple “algorithmic” approaches
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Intelligent Systems
• Perceive • Reason • Act • Communicate • Learn
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Somewhat idealized example of Infobu[ons
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Perceive the current clinical situa6on – Pa6ent characteris6cs – Clinician characteris6cs – Organiza6on characteris6cs – Clinical task – Informa6on requested
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Reason about the current situa6on – What are user needs? – What exactly is useful? – How the clinician can be helped? – What are the best resources to display? – Where to find the resources?
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Act to obtain right informa6on – Construct query – Connect to Infobu[on broker or directly to resources
– Retrieve answers
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Communicate obtained informa6on to the clinician – Aggregate informa6on from mul6ple sources – Select the most relevant informa6on – Generate natural language descrip6ons, graphics, summaries, etc.
– Display results
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Example: Infobu[ons
• Learn from experience – How results can be improved? – Get feedback if the task was successful – What worked the best and what did not work – Find out clinician preferences – Improve matching pa6ent characteris6cs to resources
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Another Example
h[p://www.aaaivideos.org/2011/robo6c_secrets_revealed_002/
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Is AI used in Healthcare?
• Most modern soYware use methods developed in AI – EMR – CDSS – Pa6ent monitoring – Scheduling & logis6cs – Mobile devices
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Why Intelligent Systems in Healthcare?
• Tradi6onal computer systems are insufficient • Need to automate/support – Tasks: Diagnoses, treatment selec6ons, … – Data: images, text, speech, …
• Need for full collabora6on between computer systems and their users
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
PROBLEMS
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems • Percep6on • Knowledge representa6on • Reasoning • Planning/decision making/problem solving • Coopera6on • Learning • Natural language/image processing • Movement & coordina6on • Crea6vity • General intelligence
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Percep6on
• The ability to sense environment and the current situa6on (“inputs” to the system)
• Coded data – Databases, forms, inputs from other systems
• Non-‐coded data (most work on percep6on) – Sensors, cameras, microphones – Object recogni6on, voice recogni6on
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Percep6on: Image Recogni6on
• The ability to recognize contents or specific features of images
• Mul6ple applica6ons – Support radiology – Computer vision – Iden6fica6on of people – Image classifica6on
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Percep6on: Image Recogni6on
• Image forma6on – Lenses, light, shading, color, …
• Image processing – Edge detec6on, texture, segmenta6on, ….
• Image/object recogni6on (very difficult) – Object appearance, seman6c recogni6on, ….
• Reconstruc6on of 3D scenes • Video processing – Object tracking, ….
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
h[p://spie.org/x8899.xml?pf=true&Ar6cleID=x8899
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Percep6on: Image Recogni6on
• Image forma6on – Lenses, light, shading, color, …
• Image processing – Edge detec6on, texture, segmenta6on, ….
• Image/object recogni6on (very difficult) – Object appearance, seman6c recogni6on, ….
• Reconstruc6on of 3D scenes • Video processing – Object tracking, …. h[p://www.youtube.com/watch?
v=1GhNXHCQGsM
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Percep6on: Natural Language Processing (NLP)
• Es6mated 70% of EHR informa6on is in notes • Different problems are being addressed – Search – Coding, informa6on retrieval – Document classifica6on – Wri6ng summaries – Transla6on – Ques6on answering
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Percep6on: Speech Recogni6on
• Transla6on of spoken words to text • Healthcare applica6ons – Automa6ng clinical documenta6on – Interac6on with computers
• NLP can be then used to code the recognized text
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Knowledge Representa6on
• The ability to store knowledge/models – Symbolic (rules, trees, ontologies) – Sub-‐symbolic (neural networks, computa6onal intelligence, evolu6onary computa6on, control theory)
– Sta6s6cal (Bayesian, HMM, SVM) – Hybrid (combina6on of the above)
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Reasoning
• Logic(s) • Algorithms (forward/backward chaining, probability upda6ng, case-‐based, …)
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Problem-‐solving
• Select the most op6mal ac6on or a sequence of ac6ons
• Search • Op6miza6on • Measuring performance – Completeness – Op6mality – Complexity (6me/space)
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Search Example: Tic-‐tac-‐toe
Spurce: wikipedia
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Planning Example • Dynamic Vehicle Rou6ng Problem (DVRP)
– Deliver goods from sources to des6na6ons
– Availability of haulage requests is not known in advance
– Global planning is infeasible
• Distributed planning approach – Agents ac6ng on behalf of single trucks in the forwarder’s transport fleet decide autonomously about handling of requests
– Agents ac6ng on behalf of single requests con6nuously valuate their own transport for the forwarder, thus seeking to ensure 6mely delivery
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Coopera6on
• Single specialized computer system is oYen not sufficient to solve complex problems
• Similarly to people, autonomous agents cooperate in solving tasks
• Examples include robo6cs, mul6agent systems, but also decision support systems, planning systems, and so on.
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Coopera6on
h[p://www.aaaivideos.org/2011/swarmanoid_the_movie
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Learning
• Learning is one of key abili6es of intelligent systems
• When interac6ng with environment intelligent systems can find ways to improve their performance
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Learning
• Supervised learning – Teacher shows how to do things, or shows examples
• Unsupervised learning – No teacher, the learner figures out that some things are similar and categorizes them
• Reinforcement learning – Learner explores to find a be[er way of doing something
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Learning
h[p://www.aaaivideos.org/2010/case_based_imita6on_sequel
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Movement
• Significant progress in AI has been made in the field of robo6cs
• Intelligent agents may take physical form, not only act as soYware agents
• Some healthcare applica6ons include – Surgery – Logis6cs & delivery – Automated pharmacy
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Movement
h[p://www.youtube.com/watch?v=vcD2leYbhxk
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Crea6vity
• What is crea6vity? • How machines can be crea6ve? • Ar6fact genera6on – Wri6ng poems, pain6ng pictures, inven6ng theories, composing music
• Three forms of crea6vity – Combinatorial – Exploratory – Transforma6onal
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Selected Problems: Ar6ficial General Intelligence
• Building of general-‐purpose “thinking machines”
• Long-‐term goal of building human-‐level intelligence
• Several projects – CAM Brain Machine (CBM)
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Strong AI vs. Weak AI
• Strong AI – building human-‐level intelligence
• Weak AI – applica6ons of AI methods
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
SELECTED METHODS/AREAS
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Mul6agent Systems
• Used to solve very large problems • Independent agents performing their tasks – Autonomy, decentraliza6on, self-‐organiza6on
• Communica6on between agents • Real-‐6me or simula6on models • Applica6ons – Transporta6on, logis6cs, planning
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Mul6agent Systems
h[p://www.youtube.com/watch?v=ilLylU1u0iQ&feature=related
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Evolu6onary Computa6on
• Class of op6miza6on methods – The goal is to find the best solu6on according to some criteria
• Used for very complex problems – In one applica6on to complex systems op6miza6on, the search space was 2.6 x 1021
– It would take 8 x 1011 years to search through all possible solu6ons (assuming 1 sec per evalua6on)
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Evolu6onary Computa6on
• The idea is to mimic evolu6onary process • Combina6on of good solu6ons is likely to produce new good solu6ons
• Uses terminology borrowed from biology – Popula6on, individual, survival, fitness, ….
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Evolu6onary Computa6on
• Many applica6ons – Management, resource alloca6on – Radiology – Treatment selec6on – Design
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
Next Week
• We will con6nue with intelligent systems and go into more of – Learning – Data Mining
• More reading on AI
HAP 752 Advanced Health Informa6on Systems Copyright © Janusz Wojtusiak, 2014
HAP 752
Janusz Wojtusiak
George Mason University [email protected]