Robust cell nuclei segmentation using statistical modelling

Abstract

The objective analysis of cytological and histological images has been the subject of research for many years. One of the most difficult fields in histological image analysis is the automated segmentation of tissue‐section images. We propose a multistage segmentation method for the isolation of cell nuclei in such images. In the first stage the compact Hough transform (CHT) is used to determine possible locations of the nuclei. We then define a likelihood function which enables us to perform an optimization procedure based on the maximization of this function. Global grey‐level histogram information is used thoughout the algorithm to reduce the amount of computation and to reject unwanted artefacts. The algorithm is tested on connective tissue images with very encouraging results. Apart from detecting well‐separated nuclei in the images, it performs well in separating dividing nuclei into likely substructures. At the same time the algorithm provides us with a measure of uncertainty along the detected boundary, in the form of the value of the likelihood function.

Automated Diagnosis of Brain Tumours Astrocytomas Using Probabilistic Neural Network Clustering and Support Vector Machines

“…Accurate segmentation of nuclei is of crucial importance to guarantee correct results in computerassisted microscopy. 26 Nuclei encode significant diagnostic and prognostic information, that if quantified can potentially allow the prediction of the disease course. Previous studies that have investigated the demanding task of nuclei segmentation [27][28][29][30] have reported relatively high segmentation accuracies, such as 85%, 27 89%, 28 and 99%.…”

Automated Diagnosis of Brain Tumours Astrocytomas Using Probabilistic Neural Network Clustering and Support Vector Machines

“…Accurate segmentation of nuclei is of crucial importance to guarantee correct results in computerassisted microscopy. 26 Nuclei encode significant diagnostic and prognostic information, that if quantified can potentially allow the prediction of the disease course. Previous studies that have investigated the demanding task of nuclei segmentation [27][28][29][30] have reported relatively high segmentation accuracies, such as 85%, 27 89%, 28 and 99%.…”

Computer-assisted three-dimensional quantitation of programmed death-ligand 1 in non-small cell lung cancer using tissue clearing technology

“…The nuclear segmentation algorithm was built based on previous methods [ 12 , 13 , 14 ]. Using the SYTO-16 channel, we applied the Li-threshold algorithm to remove background, fill foreground holes, and find connected components, obtaining rough nuclei labels.…”

Automated seeding-based nuclei segmentation in nonlinear optical microscopy

“…Therefore, a fully computerized seeding is preferable, for which several approaches are known from the literature. These approaches can be divided into three general classes: (1) local minima/maxima-based detection [8,32], (2) different variants of Hough transform [3,7], and (3) Laplacian of Gaussian (LoG) filters [4][5][6]. The local minima/maxima approaches even respond to minor intensity extrema and, therefore, are sensitive to background variations.…”

“…In common practice, the generalized version of the Hough transform is applied [45], which enables the detection of arbitrary shapes without analytic shape description. There are several variations of the Hough transform employed for nuclei detection [3,7]. In this contribution, we investigate the capability of the CHT proposed by Mouroutis et al [3] and some of its modifications to detect nuclei in NLO microscopy images.…”

“…The idea of the CHT [3] is based on the assumption that the shape of all nuclei is close to circular, and the size does not have much variation; i.e., the nuclear boundary could lie in a certain interval of distances from the centroid: from d min to d max . And vice versa, given any boundary point, there is a probability of finding the nucleus centroid at the distance from d min to d max from this boundary point (Fig.…”

Automated Diagnosis of Brain Tumours Astrocytomas Using Probabilistic Neural Network Clustering and Support Vector Machines

“…Accurate segmentation of nuclei is of crucial importance to guarantee correct results in computerassisted microscopy. 26 Nuclei encode significant diagnostic and prognostic information, that if quantified can potentially allow the prediction of the disease course. Previous studies that have investigated the demanding task of nuclei segmentation [27][28][29][30] have reported relatively high segmentation accuracies, such as 85%, 27 89%, 28 and 99%.…”

Computer-assisted three-dimensional quantitation of programmed death-ligand 1 in non-small cell lung cancer using tissue clearing technology

“…The nuclear segmentation algorithm was built based on previous methods [ 12 , 13 , 14 ]. Using the SYTO-16 channel, we applied the Li-threshold algorithm to remove background, fill foreground holes, and find connected components, obtaining rough nuclei labels.…”

Automated seeding-based nuclei segmentation in nonlinear optical microscopy

“…Therefore, a fully computerized seeding is preferable, for which several approaches are known from the literature. These approaches can be divided into three general classes: (1) local minima/maxima-based detection [8,32], (2) different variants of Hough transform [3,7], and (3) Laplacian of Gaussian (LoG) filters [4][5][6]. The local minima/maxima approaches even respond to minor intensity extrema and, therefore, are sensitive to background variations.…”

“…In common practice, the generalized version of the Hough transform is applied [45], which enables the detection of arbitrary shapes without analytic shape description. There are several variations of the Hough transform employed for nuclei detection [3,7]. In this contribution, we investigate the capability of the CHT proposed by Mouroutis et al [3] and some of its modifications to detect nuclei in NLO microscopy images.…”

“…The idea of the CHT [3] is based on the assumption that the shape of all nuclei is close to circular, and the size does not have much variation; i.e., the nuclear boundary could lie in a certain interval of distances from the centroid: from d min to d max . And vice versa, given any boundary point, there is a probability of finding the nucleus centroid at the distance from d min to d max from this boundary point (Fig.…”

Automated Diagnosis of Brain Tumours Astrocytomas Using Probabilistic Neural Network Clustering and Support Vector Machines

“…Accurate segmentation of nuclei is of crucial importance to guarantee correct results in computerassisted microscopy. 26 Nuclei encode significant diagnostic and prognostic information, that if quantified can potentially allow the prediction of the disease course. Previous studies that have investigated the demanding task of nuclei segmentation [27][28][29][30] have reported relatively high segmentation accuracies, such as 85%, 27 89%, 28 and 99%.…”

Computer-assisted three-dimensional quantitation of programmed death-ligand 1 in non-small cell lung cancer using tissue clearing technology

“…The nuclear segmentation algorithm was built based on previous methods [ 12 , 13 , 14 ]. Using the SYTO-16 channel, we applied the Li-threshold algorithm to remove background, fill foreground holes, and find connected components, obtaining rough nuclei labels.…”

Automated seeding-based nuclei segmentation in nonlinear optical microscopy

“…Therefore, a fully computerized seeding is preferable, for which several approaches are known from the literature. These approaches can be divided into three general classes: (1) local minima/maxima-based detection [8,32], (2) different variants of Hough transform [3,7], and (3) Laplacian of Gaussian (LoG) filters [4][5][6]. The local minima/maxima approaches even respond to minor intensity extrema and, therefore, are sensitive to background variations.…”

“…In common practice, the generalized version of the Hough transform is applied [45], which enables the detection of arbitrary shapes without analytic shape description. There are several variations of the Hough transform employed for nuclei detection [3,7]. In this contribution, we investigate the capability of the CHT proposed by Mouroutis et al [3] and some of its modifications to detect nuclei in NLO microscopy images.…”

“…The idea of the CHT [3] is based on the assumption that the shape of all nuclei is close to circular, and the size does not have much variation; i.e., the nuclear boundary could lie in a certain interval of distances from the centroid: from d min to d max . And vice versa, given any boundary point, there is a probability of finding the nucleus centroid at the distance from d min to d max from this boundary point (Fig.…”