Parallelization in biomedical time series analysis web platform: The MULTISAB project experience

Parallel execution of operations required for biomedical time series (BTS) analysis is an Important issue in optimization of medical software efficiency. We investigate the applicability of several parallelization approaches to BTS analysis in the context of feature extraction from multiple heterogeneous BTS on a Java-based web platform called MULTISAB, designed for medical diagnostics. Considering only the calculation parallelization of many different BTS features, our research suggests that parallelization based on simple Java multithreading works the best. The threads are assigned based on the data and analysis parameters provided, where feature extraction parallelization is performed on the following levels: 1) multiple segments; 2) multiple signal trails; 3) multiple patient records. The synchronization mechanism should be simple: the analysis continues once all threads terminate their work and record the extracted feature vectors. A special case, when features from two signal trails (bivariate features) are extracted, includes multithreading on signal pairs for multiple signal trails level parallelization. We also provide an overview of the web platform architecture to put the parallelized parts into the overall perspective.