TY - JOUR
T1 - Transition Icons for Time-Series Visualization and Exploratory Analysis
AU - Nickerson, Paul V.
AU - Baharloo, Raheleh
AU - Wanigatunga, Amal A.
AU - Manini, Todd M.
AU - Tighe, Patrick J.
AU - Rashidi, Parisa
N1 - Funding Information:
Manuscript received November 1, 2016; revised April 3, 2017; accepted May 4, 2017. Date of publication May 16, 2017; date of current version March 5, 2018. The work of P. Nickerson and P. Rashidi was supported by NIGMS under Grant 1R01GM114290. The work of A. A. Wani-gatunga was supported through a diversity supplement award by Claude D. Pepper Older Americans Independence Center NIH/NIA under Grant P30AG028740. The work of P. J. Tighe was supported by NIH NIGMS under Grant K23102697 and Grant NIGMS 1R01GM114290. Todd Manini’s work was supported by R01 AG042525, National Institute on Aging (NIA). (Corresponding author: Paul V. Nickerson.) P. V. Nickerson and P. Rashidi are with the Department of Biomedical Engineering, University of Florida, Gainesville, FL 32610 USA (e-mail: nickep@shands.ufl.edu; parisa.rashidi@ufl.edu).
PY - 2018/3
Y1 - 2018/3
N2 - The modern healthcare landscape has seen the rapid emergence of techniques and devices that temporally monitor and record physiological signals. The prevalence of time-series data within the healthcare field necessitates the development of methods that can analyze the data in order to draw meaningful conclusions. Time-series behavior is notoriously difficult to intuitively understand due to its intrinsic high-dimensionality, which is compounded in the case of analyzing groups of time series collected from different patients. Our framework, which we call transition icons, renders common patterns in a visual format useful for understanding the shared behavior within groups of time series. Transition icons are adept at detecting and displaying subtle differences and similarities, e.g., between measurements taken from patients receiving different treatment strategies or stratified by demographics. We introduce various methods that collectively allow for exploratory analysis of groups of time series, while being free of distribution assumptions and including simple heuristics for parameter determination. Our technique extracts discrete transition patterns from symbolic aggregate approXimation representations, and compiles transition frequencies into a bag of patterns constructed for each group. These transition frequencies are normalized and aligned in icon form to intuitively display the underlying patterns. We demonstrate the transition icon technique for two time-series datasets - postoperative pain scores, and hip-worn accelerometer activity counts. We believe transition icons can be an important tool for researchers approaching time-series data, as they give rich and intuitive information about collective time-series behaviors.
AB - The modern healthcare landscape has seen the rapid emergence of techniques and devices that temporally monitor and record physiological signals. The prevalence of time-series data within the healthcare field necessitates the development of methods that can analyze the data in order to draw meaningful conclusions. Time-series behavior is notoriously difficult to intuitively understand due to its intrinsic high-dimensionality, which is compounded in the case of analyzing groups of time series collected from different patients. Our framework, which we call transition icons, renders common patterns in a visual format useful for understanding the shared behavior within groups of time series. Transition icons are adept at detecting and displaying subtle differences and similarities, e.g., between measurements taken from patients receiving different treatment strategies or stratified by demographics. We introduce various methods that collectively allow for exploratory analysis of groups of time series, while being free of distribution assumptions and including simple heuristics for parameter determination. Our technique extracts discrete transition patterns from symbolic aggregate approXimation representations, and compiles transition frequencies into a bag of patterns constructed for each group. These transition frequencies are normalized and aligned in icon form to intuitively display the underlying patterns. We demonstrate the transition icon technique for two time-series datasets - postoperative pain scores, and hip-worn accelerometer activity counts. We believe transition icons can be an important tool for researchers approaching time-series data, as they give rich and intuitive information about collective time-series behaviors.
KW - Bag of patterns
KW - symbolic aggregate approximation
KW - time series
UR - http://www.scopus.com/inward/record.url?scp=85043595498&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85043595498&partnerID=8YFLogxK
U2 - 10.1109/JBHI.2017.2704608
DO - 10.1109/JBHI.2017.2704608
M3 - Article
C2 - 28534797
AN - SCOPUS:85043595498
VL - 22
SP - 623
EP - 630
JO - IEEE Journal of Biomedical and Health Informatics
JF - IEEE Journal of Biomedical and Health Informatics
SN - 2168-2194
IS - 2
ER -