Title: Use of wavelet based spectra for early detection of ovarian cancer
Authors: Dixon Vimalajeewa - Texas A&M University (United States) [presenting]
Scott Bruce - Texas A&M University (United States)
Brani Vidakovic - Texas A and M University (United States)
Abstract: Ovarian cancer presents at a late clinical-stage so that early detection of this cancer is essential for improving the survival rate. Serum mass spectrometry data collected from ovarian cancer and non-cancer patients are commonly used early diagnosis of ovarian cancer. Previous studies have mostly considered only some specific features from the spectra in detecting the presence of cancer. However, the whole spectra are accounted for and a new modality is discussed by using wavelet analysis. Wavelet analysis is a popular signal processing tool that transforms a signal into a set of coefficients representing the signal's nature at different locations and times. Wavelet spectrum is formed by using these coefficients, and the spectral slope of the wavelet spectra is used to measure the signal's regularity. The study discusses signal classification based on variability in signals regularity and its usability in the early detection of ovarian cancer. Spectral slopes of ovarian cancer spectra are computed by using wavelet spectra formed through the standard and distance covariance-based methods. Those spectral slopes are then fed into three classification algorithms, logistic regression, SVM, and KNN. Finally, the contribution of regularity in ovarian mass spectroscopy data on detecting cancer from non-cancer data is assessed with respect to the spectral slope computing methods and the classification algorithms by using correct classification rate, sensitivity, and specificity.