The article describes a novel approach to automatic recognition and classification of diseases in endoscopic videos. Current directions of research in this field are discussed. Most presented methods focus on processing single frames and do not take into consideration the temporal relationship between continuous classifications. Existing approaches that consider the temporal structure of an incoming frame sequence are focused on other domains (e.g. the segmentation of movies into scenes) and might not be able to perform well on the less diverse endoscopic videos. Therefore, a new approach is proposed, which would involve introducing two new methods in the field of automatic endoscopic video recognition. The first of the methods is the introduction of postprocessing of the results acquired by arbitrary methods of classification, taking into consideration the temporal distribution of all recognized frames in the video. The second proposes an introduction of a segmented shifting window, which would base the recognition algorithms not only on the features of the current frame, but also on multiple ones from the most recent past.The article presents the results of tests performed on a preliminary implementation of the aforementioned methods into real life endoscopic video recognition. The acquired sensitivity and specificity values are discussed and conclusions are drawn for further research.
Authors
Additional information
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2012
