Biomedical Engineering Reference
In-Depth Information
performance of their HMM system was evaluated on the two-channel QT
database in terms of waveform segmentation precision, beat detection, and
classification. Their segmentation results compared favorably to other systems
but increased beat detection with sensitivity of 99.79% and positive prediction
of 99.96% was achieved using a test set of 59 recordings. Premature ventricular
contraction (PVC) beats were also detected by using an original classification
strategy.
4.7.4 Fuzzy Methods
Fuzzy methods are frequently applied to the design of ECG-recognition sys-
tems based on rules. Features are treated as linguistic variables and given
membership values that allow a coarser classification or decision rule. A gen-
eral fuzzy system is shown in Figure 4.32 where the features are assumed
to have been preselected. Fuzzy systems differ in the fuzzy features, fuzzy
reasoning, and fuzzy rules used to obtain the diagnosis.
Xie et al. (1997) applied a fuzzy neural network (FNN) model because their
ECG features possessed numerical ranges that overlapped for both normal and
disease types. The features in their experiments consisted of physical waveform
measurements such as ST horizontal depression, ST pseudoischemic depres-
sion, and T-wave depression. The model operated on seven distinct fuzzy rules
used in combination with a standard NN output to obtain the classifier deci-
sion. Tests were conducted on a total of 42 ECG cases, of which 14 cases were
classified as normal, whereas the remaining cases were ischemia instances.
Their FNN model had a better detection rate for normal as compared to
Fuzzification of features
ECG beat feature extraction
Fuzzy reasoning using
conditional statements
Classification decision
with certainty
Fuzzy decision rules
FIGURE 4.32
General fuzzy system for detection and diagnosis of cardiovascular diseases
using ECG.
Search WWH ::
Custom Search