Abdulkadir Albayrak

Effects of color spaces to the cell segmentation in histopathological images

With the progressive development of the computer-aided diagnosis (CAD) systems, analysis of high resolution histopathological images has become more easier. In the proposed study, the effects of different color spaces used in various studies in the literature are investigated for the discrimination of cellular structures from background in histopathological images. For this purpose, performances of k-means, fuzzy c-means and expectation-maximization algorithms are compared in different color spaces. In the experimental results section, different segmentation accuracy metrics are presented in comparative manner.

Detection of mitotic cells using completed local binary pattern in histopathological images

In this study, detection of mitotic cells and the discrimination of mitotic cells from normal cells in high-resolution histopathological images are investigated. An automated model-based application tried to be developed for the detection of mitosis which is normally difficult to determine even by experts. The main purpose of this study is to assist pathologist in finding mitotic cells as second reader computer aided diagnosis system. On this purpose, firstly, k-means algorithm has been applied to distinguish the cellular structures from noncellular structures. Then, the features of this clustered cellular structures are extracted by using completed local binary pattern (CLBP). Hence, it is aimed to be sure whether the mitotic cells are able to distinguished from nonmitotic cells or not. Finally, an ensemble random Forest (RF) algorithm is used to classify the extracted features by CLBP. According to the result obtained from the study, while number of mitotic and nonmitotic cells are equal, the accuracy is significant. With increasing number of nonmitotic cells periodically cause to decrease of precision and F-measure values due to the unbalanced data distribution.

Detection of mitotic cells in histopathological images using textural features

In this work, segmentation of cellular structures in the high resolutional histopathological images and possibility of the discrimination within normal and mitotic cells has been investigated. Mitosis detection is very exhaustive and time consuming process. In the first step, features of cells which have been found by the clustering algorithm have been extracted by oriented gradient histograms (HOG) method which is known as a robust texture descriptor. A mitotic cell has some textural changes that makes it recognizable among other normal cells. Hence, the classification accuracy of the unsupervised learning methods is increased after making use of proposed textural descriptor.

Breast cancer mitosis detection in histopathological images with spatial feature extraction

In this work, cellular mitosis detection in histopathological images has been investigated. Mitosis detection is very expensive and time consuming process. Development of digital imaging in pathology has enabled reasonable and effective solution to this problem. Segmentation of digital images provides easier analysis of cell structures in histopathological data. To differentiate normal and mitotic cells in histopathological images, feature extraction step is very crucial step for the system accuracy. A mitotic cell has more distinctive textural dissimilarities than the other normal cells. Hence, it is important to incorporate spatial information in feature extraction or in post-processing steps. As a main part of this study, Haralick texture descriptor has been proposed with different spatial window sizes in RGB and La*b* color spaces. So, spatial dependencies of normal and mitotic cellular pixels can be evaluated within different pixel neighborhoods. Extracted features are compared with various sample sizes by Support Vector Machines using k-fold cross validation method. According to the represented results, it has been shown that separation accuracy on mitotic and non-mitotic cellular pixels gets better with the increasing size of spatial window.

Automatic cell segmentation in histopathological images via two-staged superpixel-based algorithms

AbstractThe analysis of cell characteristics from high-resolution digital histopathological images is the standard clinical practice for the diagnosis and prognosis of cancer. Yet, it is a rather exhausting process for pathologists to examine the cellular structures manually in this way. Automating this tedious and time-consuming process is an emerging topic of the histopathological image-processing studies in the literature. This paper presents a two-stage segmentation method to obtain cellular structures in high-dimensional histopathological images of renal cell carcinoma. First, the image is segmented to superpixels with simple linear iterative clustering (SLIC) method. Then, the obtained superpixels are clustered by the state-of-the-art clustering-based segmentation algorithms to find similar superpixels that compose the cell nuclei. Furthermore, the comparison of the global clustering-based segmentation methods and local region-based superpixel segmentation algorithms are also compared. The results show that the use of the superpixel segmentation algorithm as a pre-segmentation method improves the performance of the cell segmentation as compared to the simple single clustering-based segmentation algorithm. The true positive ratio (TPR), true negative ratio (TNR), F-measure, precision, and overlap ratio (OR) measures are utilized as segmentation performance evaluation. The computation times of the algorithms are also evaluated and presented in the study. Graphical AbstractThe visual flowchart of the proposed automatic cell segmentation in histopathological images via two-staged superpixel-based algorithms.

Superpixel approach in high resolution histopathological image segmentation

Segmentation of cellular structures with high accuracy has a crucial importance for the detection of cancerous regions in histopathologic images. The proper segmentation of cellular structures is one of the most important issues to be considered when making a diagnosis by pathologists. In this study, the contribution of the superpixel method to the segmentation of high-resolution histopathologic images of renal cell carcinoma from the TCGA (The Cancer Genome Atlas) data set was investigated. The superpixel method performs clustering based on color similarities and spatial proximity of the pixels in histopathologic images. When the results are evaluated, it has been observed that the superpixel method has a positive contribution to both the segmentation success and the running time.

Segmentation of precursor lesions in cervical cancer using convolutional neural networks

Cervical carcinoma is one of the frequently seen cancers in the world and in our country, develops from precursor lesions. These precursor lesions are analyzed by pathologists so that the diagnosis of the disease can be made. In this study, a system that performs automatic detection of pre-cancerous lesions was performed using the convolutional neural networks (CNNs). In the training phase, lesion recognition performance of the proposed system has reached 92%. Thereafter, whole image was segmented by using 60 × 60 pixel tiles during the training phase. After all, the precursor lesions were segmented with 81.71% Dice coefficient.

Superpixel-based segmentation of glottal area from videolaryngoscopy images

Abstract. Segmentation of the glottal area with high accuracy is one of the major challenges for the development of systems for computer-aided diagnosis of vocal-fold disorders. We propose a hybrid model combining conventional methods with a superpixel-based segmentation approach. We first employed a superpixel algorithm to reveal the glottal area by eliminating the local variances of pixels caused by bleedings, blood vessels, and light reflections from mucosa. Then, the glottal area was detected by exploiting a seeded region-growing algorithm in a fully automatic manner. The experiments were conducted on videolaryngoscopy images obtained from both patients having pathologic vocal folds as well as healthy subjects. Finally, the proposed hybrid approach was compared with conventional region-growing and active-contour model-based glottal area segmentation algorithms. The performance of the proposed method was evaluated in terms of segmentation accuracy and elapsed time. The F-measure, true negative rate, and dice coefficients of the hybrid method were calculated as 82%, 93%, and 82%, respectively, which are superior to the state-of-art glottal-area segmentation methods. The proposed hybrid model achieved high success rates and robustness, making it suitable for developing a computer-aided diagnosis system that can be used in clinical routines.

Evaluation of robustness of ensemble learners to noisy data

Discovering noisy data and classification of noisy data sets are problematic issues associated with noisy data sets. In our work, we used 36 UCI data sets that consist of differeent rates of noisy data to measure robustness of five ensemble learners and two basic classifiers to noisy data. According to classification success ratesof our study, Random Subspace and Bagging are more robust to noisy data than other ensemble learners and simple classifiers.