Application of complex discrete wavelet transform in classification of Doppler signals using complex-valued...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Biomedical signal classification generates large datasets, necessitating data compression for efficient processing.
• Feature extraction is crucial for reducing the dimensionality of biomedical waveform data.
• Artificial neural networks (ANNs) are widely used for biomedical signal classification but require significant computational resources.

Purpose of the Study:

• To present two novel structures, Wavelet Transform-Complex-Valued Artificial Neural Network (WT-CVANN) and Complex Wavelet Transform-Complex-Valued Artificial Neural Network (CWT-CVANN), for feature extraction.
• To reduce the size of feature sets in training and testing data for biomedical signal classification.
• To decrease the training time of artificial neural networks without sacrificing classification accuracy.

Main Methods:

• Utilized discrete real and complex wavelet transforms for feature extraction within the proposed WT-CVANN and CWT-CVANN structures.
• Applied the developed structures to classify carotid arterial Doppler ultrasound signals from patients with early atherosclerosis and healthy volunteers.
• Acquired ultrasound data from 38 patients (mean age 59) and 40 healthy individuals (mean age 23).

Main Results:

• The feature extraction algorithms successfully reduced the training times for Complex-Valued Artificial Neural Networks (CVANN) and Real-Valued Artificial Neural Networks (RVANN).
• Classification accuracy, measured by sensitivity, specificity, and average detection rate, was maintained despite the reduction in feature set size.
• The WT-CVANN and CWT-CVANN structures demonstrated efficient data compression and reduced network training duration.

Conclusions:

• Feature extraction using wavelet transforms is an effective strategy for compressing biomedical data.
• The proposed WT-CVANN and CWT-CVANN methods enable reduced training times for ANNs in signal classification tasks.
• These techniques offer a viable solution for efficient and accurate classification of complex biomedical signals like carotid Doppler ultrasound data.