Computational Physiology
Papers from the 2011 AAAI Spring Symposium
Mark Buller, Paul Cuddihy, and Finale Doshi-Velez, Program Cochairs
Technical Report SS-11-04
62 pp., $25.00
ISBN 978-1-57735-496-3
[Add to Cart] [View Cart]
The emergence of inexpensive and unobtrusive health sensors promises to shift the healthcare industry’s focus from episodic care in acute settings to early detection and longitudinal care for chronic conditions in natural living environments. The same technologies can also be used to monitor healthy individuals in high-stress work situations.
Automated human health-state monitoring aims to identify when an individual moves from a healthy to a compromised state. For example, changes in diet or physical activity can lead to life-threatening hypo or hyperglycemia in diabetics. Similarly, elderly individuals managing multiple chronic conditions may experience rapid changes in physical and cognitive health state that must be caught quickly for treatments to be most effective. Even in healthy individuals, heavy exertion in extreme climates can quickly lead to life threatening situations.
While these sensing systems are able to provide a wealth of physiological information, the noninvasive measurements are often quite different from those typically used by physicians today. The medical community is accustomed to high-quality clinical data from a limited set of sessions. Data from continuously-measuring sensors requires us to draw conclusions from large quantities of lower quality data from subacute environments where these measures are often not specific to health states of interest and can reflect the output of multiple latent variables. As the availability of longitudinal data increases, we have an unprecedented opportunity to discover new early predictors of clinically significant events.