An automatic annotation method for early esophageal cancers based on saliency guided superpixel segmentation
| Autor: | Ni Ni Rao, Xin Ming Mei, Cheng Si Luo, Yao Wen Xing, Ding Yun Liu, Tao Gan |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Superpixel segmentation
Similarity (geometry) Computer science business.industry Pattern recognition 02 engineering and technology stomatognathic diseases 03 medical and health sciences Annotation 0302 clinical medicine Manual annotation Salient 0202 electrical engineering electronic engineering information engineering 030211 gastroenterology & hepatology 020201 artificial intelligence & image processing Segmentation Computer vision Artificial intelligence Detection rate business |
| Zdroj: | Proceedings of the International Conference on Bioinformatics and Computational Intelligence. |
| Popis: | The annotation of Early Esophageal Cancer (EEC) in gastroscopic images is mainly performed by clinicians in clinic, In order to reduce subjective and fatigue manual annotation works, computer-aided annotation is needed for accurate diagnosis. This study develops an automatic annotation method to annotate EEC in gastroscopic images. We initially designed a new segmentation method named "Saliency Guided Superpixel Segmentation", to rapidly find the salient regions and their surrounding regions. Then, a two-step annotation strategy of EEC, which is based on the "relationships between the salient regions and EEC regions", was proposed. The first step is to detect the salient EEC regions and the second step is the left unsalient EEC regions. Finally, the two kinds of lesion regions are combined together to obtain the annotation results. 267 EEC images and 285 normal images were used to validate the proposed method. The experimental results show that, the detection rate and Dice similarity coefficients of our method were 97.14% and 74.53%, respectively. Compared with other patch-based or superpixel-based state-of-the-art methods, our method showed better performance. Thus, it has a good prospect for clinical application. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|