| الوصف: |
For cardiologists today, an underrated, and perhaps more precise way to detect heart defects and murmurs, is through analyzing the acoustics of the heartbeat. Rather than analyzing an electrical signal representation of the pulse, acoustic waveforms of the heart can often more directly reveal heart conditions and heart murmurs. Ultrasound Doppler is one such technique. However, noise caused by motion artifacts from body impacts during running causes a low signal-to-noise ratio. For this purpose, this paper examines the design of a Gaussian Mixture Hidden Markov model (GMHMM) that serves as a classifier of acoustic heartbeat recordings. Different feature extraction algorithms were experimented with, such as raw features, Short-Term Fourier Transform (STFT) features, wavelet transform, and finally k-means clustering, to train GMHMMs. The Expectation Maximization (EM) algorithm and the Viterbi algorithm were used to train and re-segment data respectively. |
