UCLA

This dissertation presents work in medical signal searching. Signal searching (or subsequence matching) is the process of finding similar segments within a database of time series signals. Medical signal searching is extremely challenging given the large, high entropy, multidimensional datasets. However, medical time series signals tend to be cyclical and repetitive and contain a large amount of redundancy. In addition, similar segments in a medical time series signals tend to form tight clusters where the Euclidean distance between segments within the same cluster follow a Gaussian distribution. These properties can be leveraged in many ways to significantly improve search performance in terms of computation, memory, precision, and recall.

This dissertation addresses three main aspects of medical signal searching: extraction of query segments; improvement in the quality of search results; and optimization of indexing structures through the removal of redundancy. These contributions heavily leverage the properties of medical time series signals. Each contribution of this dissertation has been thoroughly tested on a wide variety of medical time series signals including data collected through a clinical study. In addition, a complete system for searching electrocardiogram (ECG) is presented.