Summary: ECG BEATS DATABASE DESCRIPTION
by Surehka Palreddy, Ph. D. University of Wisconsin, 1996
The annotated ECG records available from the MIT/BIH (Massachusetts Institute of Technology
and Beth Israel Hospital) arrhythmia database  have been used for the evaluation of different
classifiers in this study. The database has 48 records, each 30 min in length. The data were recorded
in two channels (modified limb lead II and modified lead VI) of surface ECGs from long-term Holter
recorders. They represent a variety of waveforms, artifacts, complex ventricular, junctional, and
supraventricular arrhythmias, and conduction abnormalities. Data from 33 of the 48 records which
contain normal beats and PVCs were used for this study. Classifiers were developed and evaluated
using subsets of data from channel 1 of these 33 records sampled at 360 Hz.
Accompanying each record in the database is an annotation file in which each ECG beat has
been identified by expert cardiologists. These labels are referred to as `truth' annotations and are used
in developing the classifiers and also to evaluate the performance of the classifiers in the testing phase.
The beats in the MIT/BIH database are of several different types. In this study, we are interested
in identifying two different categories, as indicated in Table 1. Each of the two categories included
beats of several types as shown in Table 2. The AAMI  convention was used to combine the beats
into two classes of interest. The objective of this study is to determine the classification architecture
and the classification algorithm that is suitable for user adaptation of ECG beat classification systems.
Data are extracted as one feature vector for each of the beats in all the selected records. Each
vector includes one of the two possible labels according to the AAMI recommended practice. The