پایگاه داده مجاری صفرا چند بعدی
Abstract
I. Introduction
II. Related Work
III. The Multidimensional Choledoch Database
IV. How to Use the Database
V. Conclusion
Authors
Figures
References
Abstract
Histopathological examination is very important for diseases diagnosis and treatment. With the development of artificial intelligence, more and more pathological databases have been reported for histopathological diagnosis because database is quite crucial for the validation and testing of feature extraction, statistical analysis and deep learning algorithms. However, most of these databases are either gray images or RGB color images of tissue sections contain limited information of samples which limited the performance of most current deep learning algorithms. There are few publicly available pathological databases that include more than two modalities for the same subject. This paper introduces a database for both microscopy hyperspectral and color images of cholangiocarcinoma, including 880 scenes from 174 individuals, among which 689 scenes are samples with part of cancer areas, 49 scenes full of cancer areas, and 142 scenes without cancer areas. In addition, all cancer areas have been precisely labeled by experienced pathologists. The contributions of this work: a) A comprehensive and up-to-date review on pathological imaging systems and databases; b) Detailed description of the proposed the multidimensional Choledoch Database and login method; c) The multidimensional Choledoch Database has been published and can be downloaded after registration and made an entry on the website.
Introduction
Histopathological examination usiually been regarded as the ‘gold standard’ of tumor diagnosis and therapy. Traditional histopathological examination is performed by pathologist with light microscopy, which is time-consuming and laborious. With the development of image processing and artificial intelligence technology, deep learning has made great progress in pathological analysis in recent years. For example, some studies on pathological and normal voice identification [1], gastric cancer diagnosis [2], pathological retina images segmentation [3], gait analysis [4], and pathological cells recognition [5] have got high recognition accuracy. However, there still exist certain problems in the pathological diagnosis of the current artificial intelligence technology and the data resources are the most important one [6]. For most of these studies, a standard pathological database is very important to extract features, do data analysis and get the satisfied diagnosis results. In addition, the availability is also significant to guarantee the AI research results [7], [8]. So there are several researchers publishing large numbers of pathological database to support the development of algorithms such as brain database [9] and lung database [10]. In recent years, there are several latest pathological databases released for the validation and testing of feature extraction, deep learning algorithms and statistical analysis, e.g. the VOice ICar fEDerico II (VOICED) [8] and the National Cancer Data Base (NCDB) [11] and also a hyperspectral database for brain cancer [12].