You are on page 1of 1

A Situation Comparison Paradigm Based on Bio-Signals to Predict Pain Uri Kartoun, Ph.D.

ABSTRACTDuring the days following a surgery, a hospital patient often suffers from physical pain. The painkillers provided by the hospital staff may come in pill form or as injections, or be provided through a system that allows the patient to self-administer the dosage, for example, a morphine pump. This paper introduces a pain-predicting paradigm based on the collection and analysis of a large number of bio-signal readings produced by a patients body and measured in real time. These readings include measurements of skin conductivity, oxygen level, pulse rate, temperature, and dozens of other measurable bio-signals. The paradigm is based on dynamically comparing human situations. A situation is a collection of bio-signals measured during a pre-defined period (for example, 15 minutes). A baseline situation is determined as a known situation after which the patient experiences a subjectively reported, significant level of pain. The presented paradigm includes a method that dynamically measures current situations based on bio-signals and compares them to the baseline situation. A similarity rank given by the percent difference between the two situations determines how close a current situation is to the baseline situation associated with an upcoming wave of pain. Such an intelligent paradigm has the potential to notify medical officials in advance that the patient is expected to experience a significant wave of pain. Thus, medical staff can provide the patient with the appropriate pain management long before the pain starts. As a result, the patient will suffer less and will likely have a shorter stay in the hospital. In this paper, I will introduce the situation-comparison paradigm that is based on two machine-learning methods I developed. The first method, which I call signal composition, classifies time series regardless of length, type, and quantity, and the second method, which I call self-labeling supervised learning, is a supervised learning enhancement that allows the application of supervised learning methods on unlabeled training sets. The methods are demonstrated using real-time measurements of collections of time series representing financial instruments. The 8,000 time series used are financial instruments that simulate a large collection of bio-signals for a single human.

You might also like