Wireless Sensor Networks in Healthcare Presented by Mohammad Hossein Homaei President of Wireless Sensor Networks Laboratory of Iran WSNLAB Courses, 2015 Homaei@wsnlab.org Feb 2015
Potential and Challenges integrate available specialized medical tech. with wireless networks (ex: wearable accelerometers with integrated wireless cards for patient monitoring) Benefits: save on medical expenses, time (less face-to-face appointments required), allows more participants in clinical trials Homaei@wsnlab.org Feb 2015
Requirements Interoperability between biomedical devices required Event ordering, timestamps, synchronization, quick response in emergencies required Reliability and robustness for making accurate diagnoses and proper functioning in uncontrolled environments Integration of many types of sensors demands new node architecture Homaei@wsnlab.org Feb 2015
Requirements (cont.) Operation in buildings results in further interference due to walls, etc. decreasing reliability Multi-modal collaboration and energy conservation Multi-tiered data management Privacy of records: ownership of information not always clear Priority override must be carefully designed Data available during emergencies Realtime role-based access control Homaei@wsnlab.org Feb 2015
Acceptance of WSNs by patients Especially important for elderly patients: Tendency to reject technology Must be intuitive and easy to operate A study in which elderly residents of Sydney participated in an open-ended discussion found: Overall positive view of WSNs due to implications for independence Ashamed of visible sensors (design as unobtrusive as possible) Adherence issues due to forgetfulness Distrust of technology Privacy Homaei@wsnlab.org Feb 2015
Implementation Sensors: various types of wearable biomedical sensors with integrated radio transceivers (ex: accelerometer in bracelet to detect hand tremors) Ad hoc network using Zigbee protocol? Low power consumption of protocol makes it desirable for this application Radio signal received by cell phone and transmitted to server Analysis of raw data performed via wavelet analysis Decision tree or artificial neural network used to decide appropriate action (data is within normal range, outside normal range and either does or does not require emergency action, etc.) Data stored in server side database and report is generated to send to healthcare professional Homaei@wsnlab.org Feb 2015
Monitoring and Data Transmission Monitoring and transmission can occur continuously, periodically or be alert-driven (case-dependent) Transmit differential data to decrease energy consumption/traffic Priority-based transmission: path of transmission determined by nature of data, with emergency signals receiving highest priority Sensors (and potentially other wireless devices in the area) form an ad hoc network If cell phone fails to transmit data, data can be transmitted over multiple hops in ad hoc network to travel within range Homaei@wsnlab.org Feb 2015
Data Transmission (cont.) ZigBee could be appropriate specification for networking biomedical devices Significantly lower wake up time than Bluetooth (15 ms or less vs. 3 s) > low power consumption, long battery life Inexpensive transceivers Capable of establishing self-forming, self-healing mesh networks Homaei@wsnlab.org Feb 2015
Motion Detection: Wavelet Analysis Continuous Wavelet Transform (CWT)- similar to Fourier Transform, but with a variety of probing functions b translates function across x(t) and a varies time scale (t), when b=0 and a=1, represents mother wavelet of a family of wavelets problem with CWT - overly redundant and extremely difficult to recover original signal Homaei@wsnlab.org Feb 2015
Discrete Wavelet Transform To limit redundancy, DWT restricts variations in translation and scale (often to powers of two) Recovery transformation: Where a=2k, b = l * 2k, and d(k,l) is a sample of W(a,b) at discrete points Scaling function: c(n) is a series of scalars defining specific function Wavelet: d(n) is a series of scalars related to x(t) Homaei@wsnlab.org Feb 2015
Filter Banks Most basic filter bank: x(n) is divided into two - ylp(n) and yhp(n), using a digital lowpass filter H0 and highpass filter H1 respectively Homaei@wsnlab.org Feb 2015
Filter Banks (cont.) Using this method, twice the points of original function must be generated Compensate by down sampling Signal smoothed by series of low pass filters Original signal broken down into frequency bands > useful information about signal can be determined Homaei@wsnlab.org Feb 2015