Evaluation of Signal Preprocessing Methods for the Detection of Endometrial Cacer by Using near Infrared Spectroscopy

Several signal preprocessing methods used to correct near-infrared (NIR) spectra of different endometrial tissue sections have been evaluated in this paper. The real tissues sections of normal, hyperplasia, and malignant samples were used. To extract useful information and to remove the interference and background, some preprocessing methods have been compared. Particularly, spectra of the tissues section samples were assembled together to construct a 2D data matrix, so that the 2D wavelet packet transform (WPT) could be used for feature extraction. Partial least squares-discriminant analysis (PLS-DA) was used to distinguish the samples from different classes of disease states and was validated through bootstrapped Latin partition. The results of PLS-DA demonstrate that 2D WPT was the best preprocessing method among those investigated. With the decomposition level of 2 WPT, the accuracies of classification were 98 ± 2%, 99 ± 2%, and 98 ± 3%, for normal, hyperplasia, and malignant classes, respectively. The results demonstrate that NIR spectroscopy combined with 2D WPT preprocessing and proper classification methods could be a rapid, efficient, and novel means of diagnosing endometrial cancer in early stage.