Alcohol Withdrawal Syndrome Assessment based on Tremor Time-Frequency Analysi

In this thesis we established signal processing techniques to objectively evaluate the severity of Alcohol Withdrawal (AW) tremors. Many medical protocols were used pre- viously to help the physicians in assessing the severity of alcohol withdrawal, but those techniques were subjective and relied on the experience level of the physicians.;The key objective throughout this thesis is investigating the logarithmic nature of the energy emitted from tremor signals. We are able to use the energy from tremor recordings in the frequency range of [5, 15] Hz to train a logarithmic model to estimate the severity of tremors.;The next step in validating the effectiveness of the logarithmic model is the validation of the methodology in a clinical setting. The model is being validated in the emergency department for a 10-month period. During this period, each of the AW patients have been evaluated by one nurse and have been videotaped while acquiring the signal. The model provides the severity score in realtime after recording the signal. Our model is validated by comparing the score given by the model and the consensus severity rating from a panel of three expert physicians after viewing the videos. We concluded that there is a reliable agreement (kappa 0.92, 95% CI: 0.86, 0.99) between the score given by the model and the rating from our panel.;Further contributions of this thesis include an investigation of the features of AW tremors in classifying factitious vs. real tremors, based on the mean peak frequency and band-limited energy. Additionally we evaluate the differences between AW tremors in both hands and observe that by averaging the tremor ratings of each hand, a more accurate result can be obtained compared to taking either of the individual hand ratings.;Lastly, to remove the noise from the tremor signal, an Empirical Mode Decomposition (EMD) algorithm was utilized. EMD decomposes the signal into different Intrinsic Mode Functions (IMFs) and IMFs with the peak frequency in the frequency range of the tremor will be a part of the reconstructed signal. Using this technique, we successfully enhanced the accuracy of our logarithmic model.