GigaBEATS: Knowledge Discovery from Blood Pressure Data

The ALFA Group has a comprehensive effort under way to knowledge mine the arterial blood pressure signal data of ICU patients. The raw, signal level data use comes from the MIMIC II Waveform Database. Click to see our efforts in three directions and these different scalable platforms:

(*) The BeatDB Project -- The BeatDB project extracts the 1.2 billion beats of the raw signals, algorithmically defines features of these beats and compiles the transformed data into a database. Features range from being domain specific, e.g blood pressure physiology such as maximum diastole, minimum systole and beat duration, standard in signal processing, e.g based on FFTs to conventional statistics and trends such as mean arterial pressure (MAP) and its stability.

(*) The BPPR Project: Blood Pressure Pattern Recognition -- The Blood Pressure Pattern Recognition, BPPR project involves unsupervised learning, state space modeling, e.g. dynamic bayes network, and other model learning techniques adapted specially for time series pattern recognition.

(*) The UPBeat Project -- The Universal Blood Pressure Prediction Project, UPBeat involves supervised learning for time series forecasting. It includes research activities centered on the ECStar platform and the SCALE (Scalable Classification with Adaptive Learning Ensembles) system.

(*) The EC-Star Platform
(*) SCALE: Scalable Classification with Adaptive Learning Ensembles