Database Reference
In-Depth Information
eling algorithm. This simple approach proves the ability to generate
predictive alerts for hemodynamically unstable patients with high accu-
racy from trends computed on physiological signals.
In [62], a belief Bayesian belief network is developed to model ICU
data and help care givers interpret the measurements collected by patient
monitors. The belief-network model represents knowledge of pathophys-
iologic or disease states in a causal probabilistic framework. The model
is able to derive a quantitative description of the physiological states of
the patients as they progress through a disease by combining the infor-
mation from both qualitative and quantitative or numerical inputs.
Another relevant body of work on sensor mining in intensive care en-
vironments has focused on the identification and removal of undesirable
artifacts from sensor data streams. This includes mitigating the impact
of missing and noisy events, as well as clinical interventions (e.g. drawing
blood, medications) that complicate the data mining process (Section 2).
In [63], a factorial switching Kalman Filtering approach is proposed to
correct for artifacts in neonatal intensive care environments. In [64] the
authors develop clever techniques leveraging dynamic Bayesian networks
to analyze time series sensor data in the presence of such artifacts.
3.2 Sensor Data Mining in Operating Rooms
Data mining applications that relate to operating rooms tend to fo-
cus on the analysis of Electronic Medical Record data where most sensor
data inputs are filtered and summarized. For example, in [22], EMR data
is used to improve the e
ciency of operating rooms, in terms of schedul-
ing (start times, turnover times) and utilization. In [21], knowledge
management and data mining techniques are used to improve orthope-
dic operating room processes, yielding more effective decision making.
A few researchers have reported applications directly mining phys-
iological sensor data produced by operating room monitoring systems.
Exceptions are presented in [65] where the authors correlate EEG signals
with cerebral blood flow measurements for patients undergoing carotid
endarterectomy. This finding is quite valuable as it proves that EEG
signals can be used to monitor complex mechanisms including cerebral
blood flow for this patient population. In [66], machine learning tech-
niques are proposed for the closed loop control of anesthesia procedures.
In [67],the authors present a prototype of a context-aware system able
to analyze patient data streams collected in an operating room during
surgical procedures, to detect medically significant events and persist
them in specific EMR systems.
