Deep Learning-Based Earlier Detection of Esophageal Cancer Using Improved Empirical Wavelet Transform From Endoscopic Image

Abstract

In the current scenario, the research perspective on esophageal cancer becomes severe, high-prognosis malignancy; poor prognosis is primarily attributed to the fact that most tumors remain asymptomatic and unrelated before it grows through the esophagus. Significant decreases in mortality from esophageal cancer may require effective approaches to detect and nurse more patients at early, curable stages. A new Improved Empirical Wavelet Transform (IEWT) dependent on feature extraction approach and a consistent homology for the diagnosis of early esophageal endoscopic cancer have been proposed in this article. The approach is to convert an input endoscope image into CIE colored spaces L * x * y, and the x* and y* components to create a fusion image for analysis. Further, the two kinds of wavelets are obtained by adding the two forms to the fusion signal. Another is the lower-frequency component provided by the improved empirical wavelet transformation of the wave, and the other is the high- components generated from the Deep Learning-based Complex Empirical Wavelet Transformation (DL-CEWT). The fractal sizes are determined using the box interpolation method for each small block, and the abnormal regions are defined for the basis of their fractal sizes. Binary pictures are obtained by the complex threshold in each frequency variable and then divided into small blocks in every binary image. Using the homology of every block to obtain the new features in the entry image. The extraction strategies for this application are comprehensive and preliminary findings indicate that the method is effective for the early detection of esophageal cancer in an image.