|
|
 |
| |
|
|
|
A Statistical Fractal Framework for Early Detection of Cardiac Abnormalities from ECG Signals |
|
|
|
PP: 1325-1343 |
|
|
doi:10.18576/amis/200516
|
|
|
|
Author(s) |
|
|
|
Fuad S. Al-Duais,
Galal Abdulqader Ahmed Alashaari,
Yasser Mahmoud Aboelmagd,
Ahmed I. Taloba,
|
|
|
|
Abstract |
|
|
| Electrocardiogram (ECG) signals exhibit nonlinear, nonstationary, and multiscale fluctuations that reflect the underlying dynamics of cardiac electrical activity. This study proposes a statistical fractal framework for the early detection of cardiac abnormalities based on the estimation and inferential analysis of multiscale complexity measures. Wavelet Packet Decomposition is used to obtain scale-specific energy distributions, while Detrended Fluctuation Analysis (DFA), Higuchi Fractal Dimension (HFD), correlation dimension, and multifractal spectrum width are employed to quantify temporal dependence, geometric complexity, attractor structure, and scaling heterogeneity, respectively. The statistical properties of these estimators are assessed using bootstrap confidence intervals and resampling-based hypothesis tests. Differences between normal and abnormal ECG recordings are evaluated in terms of effect sizes, confidence intervals, and multiplicity-adjusted significance levels. The extracted descriptors are subsequently combined into an Adaptive Cardiac Fractal Complexity Index (ACFCI), whose weights are estimated from the training data using regularized statistical learning. The discriminative ability of the index is examined through receiver operating characteristic analysis, cross- validation, sensitivity, specificity, and calibration measures. Normal ECG signals exhibit higher fractal organization, with DFA exponents ranging from 0.98 to 1.12, HFD values from 1.55 to 1.75, and multifractal spectrum widths from 0.55 to 0.90. Abnormal recordings show lower scaling consistency, with corresponding ranges of 0.72–0.92, 1.25–1.55, and 0.20–0.55. The estimated mean correlation dimension decreases from 3.42 ± 0.18 in normal recordings to 2.17 ± 0.23 in abnormal recordings, indicating a substantial reduction in dynamical complexity. The ACFCI ranges from 0.70 to 0.90 for normal ECG signals and from 0.20 to 0.65 for abnormal signals. When incorporated into a regularized logistic classification model, the proposed index achieves an estimated cross-validated accuracy of 97.0%. Its performance is also evaluated using confidence intervals for the area under the ROC curve and other clinically relevant classification measures. These findings provide statistical evidence that cardiac abnormalities are associated with a measurable breakdown of multiscale fractal organization. The proposed framework therefore offers an interpretable inferential approach for quantifying ECG complexity and detecting cardiac abnormalities.
Keywords: |
|
|
|
|
|
|
|
