日本語フィールド
著者：Watarai, Kohei; Omura, Hajime; Minamoto, Teruya題名：Esophageal Abnormality Detection from Endoscopic Images Using DT-CDWT and Persistent Homology発表情報：Advances in Intelligent Systems and Computing 巻: 1134 ページ: 227 - 232キーワード：概要：© Springer Nature Switzerland AG 2020. We propose a new method for detecting esophageal abnormal regions from endoscopic images based on the features of the dual -tree complex discrete wavelet transform (DT-CDWT) and persistent homology. We only have to detect normal regions exactly to detect an abnormal region. More precisely, we perform two steps to detect normal regions. In the first step, we use the feature of color to detect normal regions. To this end, an input endoscopic image is converted into CIEL*a*b* color spaces, and a composite image is created from the a* and b* components. In the second step, we detect normal regions based on topological features. We divide the composite image into small blocks. We obtain the features of zero- and one-dimensional holes by applying the DT-CDWT and persistent homology to each block. We calculate the lifetime using the birth and death times of the holes. Finally, we detect the normal region from the endoscopic image based on the lifetime. We describe the proposed method in detail and the experimental results show that the method can assist doctors for endoscopic diagnoses.抄録：英語フィールド
Author：Watarai, Kohei; Omura, Hajime; Minamoto, TeruyaTitle：Esophageal Abnormality Detection from Endoscopic Images Using DT-CDWT and Persistent HomologyAnnouncement information:Advances in Intelligent Systems and Computing Vol: 1134 Page: 227 - 232An abstract：© Springer Nature Switzerland AG 2020. We propose a new method for detecting esophageal abnormal regions from endoscopic images based on the features of the dual -tree complex discrete wavelet transform (DT-CDWT) and persistent homology. We only have to detect normal regions exactly to detect an abnormal region. More precisely, we perform two steps to detect normal regions. In the first step, we use the feature of color to detect normal regions. To this end, an input endoscopic image is converted into CIEL*a*b* color spaces, and a composite image is created from the a* and b* components. In the second step, we detect normal regions based on topological features. We divide the composite image into small blocks. We obtain the features of zero- and one-dimensional holes by applying the DT-CDWT and persistent homology to each block. We calculate the lifetime using the birth and death times of the holes. Finally, we detect the normal region from the endoscopic image based on the lifetime. We describe the proposed method in detail and the experimental results show that the method can assist doctors for endoscopic diagnoses.