Efficient nucleus detector in histopathology images

Abstract

SummaryIn traditional cancer diagnosis, (histo)pathological images of biopsy samples are visually analysed by pathologists. However, this judgment is subjective and leads to variability among pathologists. Digital scanners may enable automated objective assessment, improved quality and reduced throughput time. Nucleus detection is seen as the corner stone for a range of applications in automated assessment of (histo)pathological images.In this paper, we propose an efficient nucleus detector designed with machine learning. We applied colour deconvolution to reconstruct each applied stain. Next, we constructed a large feature set and modified AdaBoost to create two detectors, focused on different characteristics in appearance of nuclei. The proposed modification of AdaBoost enables inclusion of the computational cost of each feature during selection, thus improving the computational efficiency of the resulting detectors. The outputs of the two detectors are merged by a globally optimal active contour algorithm to refine the border of the detected nuclei. With a detection rate of 95% (on average 58 incorrectly found objects per field-of-view) based on 51 field-of-view images of Her2 immunohistochemistry stained breast tissue and a complete analysis in 1 s per field-of-view, our nucleus detector shows good performance and could enable a range of applications in automated assessment of (histo)pathological images.

Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images

“…Ali et al [18] proposed an active contour-based approach to detect and segment overlapping nuclei based on shape models, which is highly variable in the case of tumor nuclei. In a recent study, Vink et al [19] employed AdaBoost classifier to train two detectors, one using pixel-based features and the other based on Haar-like features, and merged the results of two detectors to detect the nuclei in immunohistochemistry stained breast tissue images. The performance of the method was found to be limited when detecting thin fibroblasts and small nuclei.…”

Computer-aided prognosis on breast cancer with hematoxylin and eosin histopathology images: A review

“…19 Recently, various nuclei detection and segmentation approaches have been put forward for H&E histopathology images of BC (Table 1). Nuclei detection algorithms consist of voting-based, 54 Laplacian of Gaussian (LoG)-filter based, 55 intensity-based, 56 mathematical morphologybased, 57,58 H-minima transform-based, 59 watershedbased, [60][61][62] gradient-based, 58 color-based, 63 region growth and Markov random field (MRF), 64 Gaussian mixture model, 65 and deep learning 41 approaches. Although those methods show efficiency in nuclei detection, finding proper seed points or deciding initial contours is still very challenging for H&E images.…”

Dual-Channel Active Contour Model for Megakaryocytic Cell Segmentation in Bone Marrow Trephine Histology Images

“…Within the context of cell segmentation in pathology images, most of the recently proposed methods focus on nuclei segmentation as cytoplasm segmentation remains a challenging task in many types of tissue images [10], [13], [14], [39]. For instance, Ali et al propose an adaptive active contour model with shape prior for nuclear segmentation in prostate cancer tissue [40].…”

“…Image segmentation approaches based on deformable models, such as level set methods, have been widely used to successfully delineate various structures in pathology images [10]- [14] and widely applied for tissue and nucleus segmentation for cancer diagnosis, such as breast cancer [11]- [13], [15]- [17] and prostate cancer [18], [19]. In a more general context, region-based level set methods [20]- [24] have been particularly successful by incorporating region-based statistical information into an energy functional.…”