Szczegóły publikacji
Opis bibliograficzny
Color normalization approach to adjust nuclei segmentation in images of hematoxylin and eosin stained tissue / Adam PIÓRKOWSKI, Arkadiusz Gertych // W: Information Technology in Biomedicine : 6th international conference, ITIB'2018 : Kamień Śląski, Poland, June 18–20, 2018 : proceedings / eds. Ewa Pietka, [et al.]. — Cham: Springer International Publishing AG, cop. 2019. — (Advances in Intelligent Systems and Computing ; ISSN 2194-5357 ; vol. 762). — ISBN: 978-3-319-91210-3; e-ISBN: 978-3-319-91211-0. — S. 393–406. — Bibliogr. s. 405–406, Abstr. — Publikacja dostępna online od: 2018-06-06
Autorzy (2)
- AGHPiórkowski Adam
- Gertych Arkadiusz
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 115331 |
|---|---|
| Data dodania do BaDAP | 2018-07-31 |
| DOI | 10.1007/978-3-319-91211-0_35 |
| Rok publikacji | 2019 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Konferencja | 6th international conference on Information Technologies in Biomedicine |
| Czasopismo/seria | Advances in Intelligent Systems and Computing |
Abstract
Lack of standards in hematoxylin and eosin (H&E) tissue staining across laboratories is one of the reasons for differences in appearance of specimens under the microscope. It also negatively impacts the performance of digital image analysis algorithms, including nuclei segmentation that is deemed to be affected the most. To alleviate this problem, we searched through the color space to find color targets to which coloration of the original H&E image can be transferred with the goal to improve performance of a baseline nuclear segmentation method. Color targets that we found were plugged into the Reinhard’s color normalization algorithm to transfer the original H&E image to a new color space. The color-transferred images were then processed by two proposed approaches that subtract and subsequently threshold red and blue color channels. Implementation of these steps improved the amount of false positive pixels and splitting of clustered nuclei in the nuclear mask generated by the baseline method. The pixel-based segmentation accuracy was 94% in selected images. The performance was assessed in heterogeneous images of colon with manually delineated nuclei.
