Time-Series Data Analysis in Biomedical Applications

Abstract

This thesis explores methods for analyzing biomedical time-series data, focusing on two distinct applications: audio-based cough detection and time-gate optimization in Fluorescence Lifetime Imaging Microscopy (FLIM).

The first section presents time-gate optimization algorithm for rapid lifetime determination (RLD) in FLIM applications. FLIM is an emerging imaging technique used to measure molecular interactions in biological samples. The developed algorithm focuses on optimizing the time gates to balance speed and accuracy, which is particularly beneficial under diverse noise conditions. By maximizing the signal-to noise ratio (SNR), the algorithm improves the precision of lifetime measurements, enabling efficient analysis of biological processes that require fast imaging rates, such as cellular metabolism and neurological activities.

The second section presents a machine learning algorithm for automated cough detection using Convolutional Recurrent Neural Networks (CRNNs). Leveraging advanced feature extraction techniques, such as Mel spectrograms, the algorithm effectively distinguishes cough events from other audio signals, achieving high accuracy. Its adaptability to varying noise conditions makes it ideal for real-time respiratory monitoring, with strong potential for integration into mobile health platforms and hospital systems. This work addresses the critical need for non-invasive, continuous monitoring tools for chronic cough, a condition that significantly affects quality of

life.

Both contributions highlight the potential of targeted time-series analysis to improve the accuracy, speed, and reliability of biomedical monitoring and imaging. By advancing methods for cough detection and fluorescence lifetime estimation, this thesis offers adaptable tools for broader biomedical applications, contributing to both healthcare diagnostics and biological research.