Toward_Next_Generation_Health_Assistants

Download Report

Transcript Toward_Next_Generation_Health_Assistants

Objectives
Applications
• Use semantic technologies to encode and integrate a wide range
of health information to help people function at a level higher than
their training.
Towards Next Generation Integrative Mobile
Semantic Health Information Assistants
Evan W. Patton ([email protected]), John Sheehan ([email protected]),
Yue Robin Liu ([email protected]), Bassem Makni ([email protected]), and
Deborah L. McGuinness ([email protected])
Rensselaer Polytechnic Institute / 110
8th
Street / Troy, NY, 12180 USA
http://tw.rpi.edu/web/project/MobileHealth
Use Case
A patient observes from her blood report that her Blood Urea Nitrogen has
trended above the recommended range. She asks her oncology nurse about
this phenomenon and what it might mean for her treatment. The nurse informs
her that this is a common occurrence and usually due to dehydration. The
patient makes a note in her personal health assistant to track this information
further, and the application reminds her that she has a scale at home that can
measure water mass in the body as a percentage of overall mass and that her
MyFitnessPal application allows her to track her water intake. Using a sensor
and an app, the patient collects her own data and becomes more active in
monitoring her health. Further, she notices that her blood glucose is higher
than normal. Using a BlueTooth™ enabled glucose meter, she takes additional
readings over the course of the day that are synchronized to her smartphone
and made available as part of an augmented electronic health record.
Clinical Notes
Ontological Resources
Online Resources
Textual (PDF, JPEG),
Audio (WAV, MP3, AIF)
SNOMED-CT, UMLS,
HL7, Semantic Sensors, BioPortal
WebMD, PubMed
DrugBank, DBPedia,
Cancer Forums/Wikis
converted via
structured terms, relationships
unstructured or
semistructured text
Open Source
OCR/SR Software
Tesseract / OCRopus,
Sphinx / Julius
Accelerometer
natural
language
content extraction / matching
Scale
Medinet
Health Apps /
Medical Devices
capture data
User
pose questions
view data
record data
Structured representation of global
knowledge and personal health data
Provenance / trust annotations about
data sources and processing
Electronic
Medical Records
Heart Rate
MediNet is joint work with Heng Ji.
Sleep
queries graph
share data
Smart Phone
Device
share data
submit queries
answer queries
Watson-like
Question
Answering
Benefits of Semantics
By integrating data and providing a semantic, machinereadable and understandable format for health data, we
can take advantage of tools such as IBM’s Watson Q&A
system by providing its Type Coercion (TyCor) system [1]
with relevant class descriptions (either those explicitly
declared in existing ontologies or discovered through
subsumption inference) of the end user’s medical data and
match it to unstructured information from various healthrelated sources such as the CDC, John Hopkins, the Mayo
Clinic, WebMD, and the National Institutes of Health.
Data Products
The result of all the mobile health monitoring efforts are aggregated into a
structure we call Medinet. Medinet combines structured health and fitness data
with natural language processing techniques on clinical notes, online forums, and
medical literature to provide a homogenized representation for all the various
types of data and knowledge gathered by out integration efforts.
User Device
PHR/EHR
Physician
Conclusions
Data Interfaces
Acknowledgments:
Blood Pressure
Pedometers
Fact Extraction /
Cross-Source
Truth Finding
National Science Foundation
RPI – Rensselaer Polytechnic Institute
TWC – Tetherless World Constellation at Rensselaer Polytechnic Institute
EHR / EMR – Electronic Health Record / Electronic Medical Record
PHR – Personal Health Record
CDC – Center for Disease Control
Hardware Abstraction Layer / Device APIs
PHR/EHR
Sponsors:
Glossary:
Mobile Semantic Health Integration Framework
• Leverage existing curated and uncurated sources, build reusable
integrated content sources and infrastructure
Introduction
We aim to build a 24x7 health assistant platform that uses semantic
technologies to integrate health data from a variety of resources to support
patients in understanding when and how to actively engage in their care,
thus potentially increasing a feeling of informed control and empowerment.
We motivate the work with a use case helping a newly diagnosed breast
cancer patient understand and participate in her treatment plan. We crossreference data and information in clinical notes with broader medical online
resources using natural language processing (e.g., Cross-source
Information Extraction, Text Mining and Linking) and Watson-inspired
technologies to provide contextualized and referenced information to
support browsing and question answering.
Reasoning Services
Information are then presented in a variety
of web and mobile interfaces to allow
patients instant access to their health data.
Further, we can use device information,
such as location and calendar, to augment
the health data in our knowledge base.
Mobile devices and sensors enable new opportunities to empower people
to be more proactive about their healthcare and as a means to disseminate
personalized information using rich data sources integrated from across the
web. Further, advances in question answering systems such as IBM’s
Watson provide a vision where smart assistants provide targeted answers
based on an individual’s circumstances.
Our work makes available many different resources as structured data that
can be queried and integrated into new software systems to address the
needs of patients. We have prototyped mobile applications to deliver this
information along with inferences on whether measurements are within
normal ranges and are continuing to explore sensor integration abilities.