A Novel Probabilistic Nuclei Segmentation Algorithm for H&E Stained Histopathological Tissue Images

Faruk SERİN; Metin ERTÜRKLER; Mehmet GÜL

doi:https://doi.org/10.2339/politeknik.464541

Makale detayları

A Novel Probabilistic Nuclei Segmentation Algorithm for H&E Stained Histopathological Tissue Images

Yazarlar:

Faruk SERİN | Metin ERTÜRKLER | Mehmet GÜL

Yayıncı:

Gazi University

Ana Başlıklar:

Uygulamalı fizik

Dergi:

Journal of Polytechnic 2020 Vol.23 Issue 1, pp.7-17

Belge Tipi:

Makale

Yayın Tarihi:

2/2020

Belge Dili:

İngilizce

Document ID (Org No.):

2000589786

Sıklık:

Aylık

eISSN:

2147-9429

DOI:

https://doi.org/10.2339/politeknik.464541

Özet:

 












In this study, we propose a novel, fast and accurate
segmentation algorithm to segment nuclei in H&E stained histopathological
tissue images. The proposed algorithm doesn’t require pre-processing,
post-processing, and any manual parameter or threshold. The algorithm utilizes
probabilistic and statistical properties of the pixels’ color value in the
images with RGB color, and determines whether pixels are a part of any nuclei
or not by using an automatically calculated threshold value. The algorithm
provides time efficiency and reduced overall cost in the segmentation. The
other contributions of the study are false positive removal algorithm and
automatically determination of nuclei cluster for K-means. In order to compare
and evaluate the performance of the proposed algorithm in terms of time and
cost efficiency, K-Means is preferred because of its common usage. Expert
evaluation is declared as ground truth for determining the accuracy of the
results. The experiments are performed on 60 healthy and 60 damaged kidney, and
60 healthy and 60 damaged liver tissue images. The evaluations are revealed
that the proposed algorithm can effectively segment nuclei. The comparison
results also demonstrate that the deviation between proposed algorithm and the
expert is 2%, while the deviation between K-Means and Expert is 5%.


.

 


In
this study, we propose a novel, fast and accurate segmentation algorithm to
segment nuclei in H&E stained histopathological tissue images. The proposed
algorithm does not require pre-processing, post-processing, and any manual
parameter or threshold. The algorithm utilizes probabilistic and statistical
properties of the pixels’ color value in the images with RGB color space, and
determines whether pixels are a part of any nuclei or not by using an
automatically calculated threshold value. The algorithm provides time
efficiency and reduced overall cost in the segmentation. Two more algorithms
are also proposed to distinguish nuclei cluster from the other clusters
obtained by K-means, and eliminate false positives in nuclei cluster, which are
not nuclei. In order to compare and evaluate the performance of the proposed
segmentation algorithm in terms of time and cost efficiency, K-Means is
preferred because of its common usage. Expert evaluation is declared as ground
truth for determining the accuracy of the results. The experiments are
performed on 60 healthy and 60 damaged kidney, and 60 healthy and 60 damaged
liver tissue images. The evaluations show that the proposed algorithm can
effectively segment nuclei. The comparison results also demonstrate that the
deviation between proposed algorithm and the expert is 2%, while the deviation
between K-Means and expert is 5%.


.

Makale detayları

A Novel Probabilistic Nuclei Segmentation Algorithm for H&E Stained Histopathological Tissue Images

Yazarlar: Faruk SERİN | Metin ERTÜRKLER | Mehmet GÜL

Yayıncı: Gazi University

Ana Başlıklar: Uygulamalı fizik

Dergi: Journal of Polytechnic 2020 Vol.23 Issue 1, pp.7-17

Belge Tipi: Makale

Yayın Tarihi: 2/2020

Belge Dili: İngilizce

Document ID (Org No.): 2000589786

Sıklık: Aylık

eISSN: 2147-9429

DOI: https://doi.org/10.2339/politeknik.464541

Sayı

Yazarlar:

Faruk SERİN | Metin ERTÜRKLER | Mehmet GÜL

Yayıncı:

Gazi University

Ana Başlıklar:

Uygulamalı fizik

Dergi:

Journal of Polytechnic 2020 Vol.23 Issue 1, pp.7-17

Belge Tipi:

Makale

Yayın Tarihi:

2/2020

Belge Dili:

İngilizce

Document ID (Org No.):

2000589786

Sıklık:

Aylık

eISSN:

2147-9429

DOI:

https://doi.org/10.2339/politeknik.464541