Feature Extraction Based on the Wavelets and Persistent Homology for Early Esophageal Cancer Detection From Endoscopic Image

A new feature extraction method based on the wavelets and persistent homology for early esophageal cancer detection from an endoscopic image is proposed. In our proposed method, an input endoscopic image is converted to CIE L*a*b* color spaces, and a fusion image is made from the a* and b* components. Applying the two types of wavelets to the fusion image, the two types of frequency components are obtained. One is the low frequency component obtained by the dyadic wavelet transform (DYWT), and the other is the high frequency components obtained by the dual-tree complex discrete wavelet transform (DT-CDWT). Applying the dynamic threshold to each frequency component, binary images are obtained, and then each binary image is divided into small blocks. Utilizing the persistent homology to each block, the new features of the input image are acquired. The method to extract the feature is described in detail, and experimental results are presented to demonstrate that our method is useful for the development of early esophageal cancer detection from endoscopic image.