Speech recognition methods applied to biomedical signals processing.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Data Science

Background:

• Long biological signals from monitoring devices like ECG, ICP, and EEG are crucial for diagnosing various conditions.
• Traditional analysis methods can be computationally intensive and may not capture the complex temporal dynamics of these signals.

Purpose of the Study:

• To adapt and evaluate speech processing techniques, specifically Dynamic Time Warping (DTW) and Hidden Markov Models (HMM), for the analysis of long biological signals.
• To compare the efficacy of DTW and HMM in processing electrocardiogram (ECG) for arrhythmia analysis, intracranial pressure (ICP) monitoring, and electroencephalogram (EEG) for sleep apnea detection.

Main Methods:

• Application of Dynamic Time Warping (DTW) for unsupervised analysis of ECG and ICP signals using hierarchical clustering.
• Utilizing Hidden Markov Models (HMM) for sleep stage estimation and breathing event detection in EEG signals via Viterbi inference.

Main Results:

• DTW demonstrated superior performance over HMM in the analysis of ECG and ICP signals.
• HMM, leveraging its inference capabilities, proved effective for diagnosing sleep apnea from EEG data.

Conclusions:

• Speech processing techniques offer viable solutions for analyzing complex biological signals.
• The choice of method (DTW or HMM) depends on the specific application and signal characteristics for optimal diagnostic accuracy.