Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
A Physiologically-Aware Architecture for Fidelity-Preserving Transmission of Biomedical Signals in Body Area Sensor Networks and IoT
Abstract
This dissertation research presents a novel physiologically-aware and fidelity-preserving communication architecture for the transmission of biomedical signals for applications in Wireless Body Area Sensor Networks (WBASNs), Wearables and Medical Internet of Things (MIoT). This architecture is highly suitable in the current emerging landscape where machine-learning (ML) algorithms are being increasingly relied upon to supplement and enhance medical expertise for telemonitoring and e-health applications. With the inclusion of ML components, the architecture leverages opportunities for very high energy savings by taking advantage of a notion we refer to as the signal’s ‘classifiability’. Classifiability refers to the ability of suitably-chosen ML algorithms to successfully perform necessary diagnostic-related classification tasks in the face of—and specifically resulting from—compressor-induced distortion. Experimental research results show that the classifiability’s rate-distortion latitude when viewed from a different space (called ‘feature space’) is much wider than the traditional time-domain (sample space) rate-distortion latitudes. The importance of distortion measures, both in sample space and feature space for reconstruction-error evaluation are emphasized. In particular, a reconstruction-error distance metric called the diagnostic distortion measure (DDM) is also presented as a fast, and straightforward measure for evaluating distortion in feature space for implementation in resource-constrained embedded systems. Because biomedical signals carry important diagnostic information that is of clinical significance for diagnosis, our research results demonstrate how the integration of distortion measures in sample space as well as feature space can be used jointly or in isolation to ensure the features of clinical significance are duly protected and their reconstruction guaranteed. The feature-extraction capabilities of the architecture enable on-device pre-screening tasks of the subject being monitored and enable the detection of significant diagnostic events (SDEs) by assigning the event a symptomatic abnormality indicator score (SAI) that can be used to modulate parameters of the architecture’s transmission pipeline, and controlling the energy-expenditure capability as dictated by the distortion-rate latitude in the classifiability scale. This endows the architecture with a ‘physiologically-aware’ component that is uniquely tied to this research work. Joint accuracy-distortion-rate analysis is performed using publicly available datasets for EEG and ECG to fully demonstrate the viability of the proposed architecture across the two most frequently utilized biomedical signal classes.
Subject Area
Computer Engineering|Electrical engineering|Computer science
Recommended Citation
Santos, Jose M, "A Physiologically-Aware Architecture for Fidelity-Preserving Transmission of Biomedical Signals in Body Area Sensor Networks and IoT" (2020). ETD collection for University of Nebraska-Lincoln. AAI28085788.
https://digitalcommons.unl.edu/dissertations/AAI28085788