Transcript presentation - Computer Science, Stony Brook University

AnamneVis: A Framework for the
Visualization of Patient History and
Medical Diagnostics Chains
Zhiyuan Zhang1, Faisal Ahmed1, Arunesh Mittal1,
IV Ramakrishnan1, Rong Zhao1, Asa Viccellio2, Klaus Mueller1
1
2
Department of Computer Science
Department of Emergeny Medicine
Medical History - Anamnesis
Includes
• patient demographics, problems, symptoms, diagnoses, progress
notes, treatments, medication, vital signs, past medical history,
immunizations, laboratory data, radiology reports, and many
others.
Should:
• accurately reflect the diseases/problems
• indicate the probable cause of disease
Paper-Based Medical Record
Electronic Medical Record (EMR)
• Full of documentation + tables
• Non-intuitive interfaces
• Fragmented display of patient information
Problems of Current EMR
Acceptance of the EMR in clinical practice lags far behind
its expectation and potential.
Related information and overviews are difficult to obtain
•
Severely impedes a physician’s diagnostic reasoning.
Possible reason:
• Inefficient, fragmented display of patient information
Objectives
•
Simple Interfaces to make data and information exploration
easier
•
Comprehensibly organizing the patient medical history from
different sources
•
Support and enhance the clinical decision-making
• Visualization and interaction are the key
• Provide a suitable visual mapping for each visual information.
• Ease of data and information access is the key
• AnamneVis – Visualization of Anamnesis
System Overview
The 5-W Scheme
• Use a strongly structured paradigm , the 5-W
• WHO and WHAT :
• the patient and the history, in terms of symptoms, tests
and results, diagnosis, treatments and medications, etc.
• ICD, ICD-procedure.
• WHERE : locations (when appropriate) of the WHAT on the
human body
• WHEN , WHY and HOW : show a case under (doctor)
collaborative diagnosis/treatment, or an entire life span. It
demonstrates for each node what, when, why, and how that
node appears.
Identify the Five W’s :
Information Extraction
Input:
• Medical reports
• Doctor-patient dialogs and other interactive inputs
• Results from triage, and data acquired from the patient, such as
radiological images, lab analyses...
NLP (Natural Language Processing) engine (cTake)
+ online medical ontology server(SNOMED)
• Extract structured information and relationships.
• Formats the extraction results into the Five W model
• Passes it on to the visualization engine.
Visual Encoding of 5-W’s
Two cooperating displays:
• A hierarchical radial (patient overview) display with an integrated
body outline primarily for the who and where.
• A sequential (diagnostic reasoning) display primarily for the when,
why, and how.
Visual Encoding of 5-W’s
Two cooperating displays:
• A hierarchical radial (patient overview) display with an integrated
body outline primarily for the who and where.
• A sequential (diagnostic reasoning) display primarily for the when,
why, and how.
The what is part of both displays
• In form of the various nodes
•
Context-sensitive.
The two interfaces are linked
• operations on either view will be reflected in the other.
Hierarchical Radial Display – Data Model
Data Model:
• Tree data structure to store the code hierarchy information.
• Insert each incident into the tree
Node structure:
Hierarchical Radial Display – Visual Design
Tree Visualization.
• Node+Edge
• Hard to show additional info. for each node and integrate body map
Hierarchical Radial Display – Visual Design
Tree Visualization.
• Space filling:
•
Circle Pack: Size grows very fast.
•
Icicle: Not fully utilize space; Difficult to integrate body map
Hierarchical Radial Display – Visual Design
We use the sunburst visualization paradigm.
• Space-filling diagram.
• Fully utilize the space.
• Easy to integrate with a body map.
Hierarchical Display Examples
Hierarchy-centric
• Node is sized by how many sub-categories it has
• Focus more on the hierarchical information of the medical
codes
• Serve as an illustration of the complexity of a sub-system and
its composition
Hierarchical Display Examples
Patient-centric
• More radial space is dedicated for diagnoses/procedures the
patient had activities in.
• Other nodes will be collapsed to save space for others
Hierarchical Display Examples
Three predefined code hierarchies to support multi-scale
visualization.
•
Level 1: the highest code hierarchy level.
•
Level 2: more detailed categories.
•
Level 3: incident nodes (symptoms/procedures/diagnosis that
the patient has activities in. )
Help users quickly explore these three levels.
Users can expand and collapse the nodes interactively by
their expertise.
Typical Diagnostic Information Flow
Patient visits
physician
Obtain
patient’s
information
Give
treatments
Physical
examination
Make
diagnosis
Order
laboratory
data
Request and
obtain
consultation
A sequential display can show these reasoning chains very well.
Sequential (Causal) Display
Demonstrate the what, when, why and how information.
The medical records are organized by an underlying graph
data structure.
Each node corresponds to one incident (medical primitive),
which could be a doctor visit, symptom, test/data,
diagnosis or treatment.
Edges represent relationships.
Visual Design
A node is displayed as one elongated box Diagnosis
• Better utilizes the rectangular screen
• Better fits the text
Diagnosis
• Better scalability
Edges represent relationships.
Edge bundling is used to reduce cluttering.
Back edges
• May be due to treatments causing new symptoms, or
treatments constituting doctor referrals.
• Shown in different color (red) to make them easy to see.
Sequential Display Example
Sequential Display Example
Conclusions
Implemented a preliminary framework for clinical visual
analytics
• For clinical scenarios
• Patient-focused
• Unifies all EMR information fragments into a single interactive
visual framework
Current/Future Work
Improvements, fine-tuning, and additional features
• Include more formal user and affordance studies
• Potentially used by coding personnel who work in the
hospital billing office to translate medical records to ICD
code.
- Better recognize relationships in medical services.
- Perform more accurate billing statements.
• Treatment Outcome
• Google body browser
• Integrate with current EMR system.
Thanks For Listening!
Q&A?