Automatic Ki-67 Counting Using Robust Cell Detection and Online Dictionary Learning

Abstract

Ki-67 proliferation index is a valid and important biomarker to gauge neuroendocrine tumor (NET) cell progression within the gastrointestinal tract and pancreas. Automatic Ki-67 assessment is very challenging due to complex variations of cell characteristics. In this paper, we propose an integrated learning-based framework for accurate automatic Ki-67 counting for NET. The main contributions of our method are: 1) A robust cell counting and boundary delineation algorithm that is designed to localize both tumor and nontumor cells. 2) A novel online sparse dictionary learning method to select a set of representative training samples. 3) An automated framework that is used to differentiate tumor from nontumor cells (such as lymphocytes) and immunopositive from immunonegative tumor cells for the assessment of Ki-67 proliferation index. The proposed method has been extensively tested using 46 NET cases. The performance is compared with pathologists' manual annotations. The automatic Ki-67 counting is quite accurate compared with pathologists' manual annotations. This is much more accurate than existing methods.

An Advanced Deep Learning Approach for Ki-67 Stained Hotspot Detection and Proliferation Rate Scoring for Prognostic Evaluation of Breast Cancer

“…The automatic scoring will provide high throughput, more objective and reproducible results in comparison with the manual evaluation. The proliferation score is calculated as the ratio between total numbers of immunopositive nuclei and a total number of nuclei present in the image 11 . The immunopositive (brown color) and immunonegative (blue color) nuclei together called as hotspots.…”

Robust Nucleus/Cell Detection and Segmentation in Digital Pathology and Microscopy Images: A Comprehensive Review

“…The marker detection results using this method on several sample images are shown in Figure 1. Considering the nucleus scale variations, a single-pass voting with multiple scales is presented for nucleus locating in H&E stained lung cancer images [118], [119], [120], [81] and IHC stained pancreatic neuroendocrine tumor (NET) images [121], [82]. In [83], a similar single-pass voting in applied to only touching or overlapping nuclei which are discriminated from those isolated nuclei by using ellipse descriptor analysis, and thus reduces the number of voting points.…”

“…In [196], [250], [251], the GVF snake is applied to the LUV color space for simultaneous segmentation of nuclei and cytoplasm in blood smear images; in [252], a repulsive force is incorporated to the GVF snake to separate adjacent neuronal axons, which is achieved by reversing the gradient direction of neighboring objects. Recently, Xing et al [121], [82] have introduced a contour-based repulsive term into the balloon snake model [242] for nucleus segmentation in pancreatic neuroendocrine tumor images. The internal F int and external F ext forces of the repulsive balloon snake model are represented as $$\begin{array}{cc}right{F}_{int}\left({\mathrm{bold-italicv}}_i\right)& left=\alpha {\mathit{v}}_i^{″}\left(s\right)-\beta {\mathit{v}}_i^{″″}\left(s\right),\\ right{F}_{ext}\left({\mathrm{bold-italicv}}_i\right)& left=\gamma {\mathit{n}}_i\left(s\right)-\lambda \frac{\nabla E_{ext}\left({\mathrm{bold-italicv}}_i\left(s\right)\right)}{\Vert E_{ext}\left({\mathrm{bold-italicv}}_i\left(s\right)\right)\Vert }\end{array}$$$$+\omega \sum _{j=1,j\ne i}^N{\int }_0^1{d}_{ij}^{-2}(s,t){\mathrm{bold-italicn}}_i\left(t\right)dt,$$ where the two terms in (29) are the second and fourth derivative of v i ( s ) with corresponding weights α and β , respectively.…”

“…The segmentation results using [82] on one sample pancreatic neuroendocrine tumor image. Red dots and contours represent results of marker detection and boundary segmentation in (b) and (c), respectively.…”

Automated Ki-67 Quantification of Immunohistochemical Staining Image of Human Nasopharyngeal Carcinoma Xenografts

“…In which pixel intensity thresholding methods67 were to make use of pixel intensity in red, green and blue (RGB) color space, and applied intensity transformation and global thresholding according to differences between colors of brown and blue. Edge-based methods8910 were to make use of pixel intensity, gradient flow or other characteristic morphological differences between both sides of the cell boundaries for segmentation, on which rely to look for boundaries. While classification methods took the single pixel as the object of study and pixels in the same category together constitute each component of tissues, in which both supervised11 and unsupervised12 learning approaches have been applied with the difference that whether training samples are needed.…”