An optimized image analysis algorithm for detecting nuclear signals in digital whole slides for histopathology

Abstract

Nuclear estrogen receptor (ER), progesterone receptor (PR) and Ki-67 protein positive tumor cell fractions are semiquantitatively assessed in breast cancer for prognostic and predictive purposes. These biomarkers are usually revealed using immunoperoxidase methods resulting in diverse signal intensity and frequent inhomogeneity in tumor cell nuclei, which are routinely scored and interpreted by a pathologist during conventional light-microscopic examination. In the last decade digital pathology-based whole slide scanning and image analysis algorithms have shown tremendous development to support pathologists in this diagnostic process, which can directly influence patient selection for targeted- and chemotherapy. We have developed an image analysis algorithm optimized for whole slide quantification of nuclear immunostaining signals of ER, PR, and Ki-67 proteins in breast cancers. In this study, we tested the consistency and reliability of this system both in a series of brightfield and DAPI stained fluorescent samples. Our method allows the separation of overlapping cells and signals, reliable detection of vesicular nuclei and background compensation, especially in FISH stained slides. Detection accuracy and the processing speeds were validated in routinely immunostained breast cancer samples of varying reaction intensities and image qualities. Our technique supported automated nuclear signal detection with excellent efficacy: Precision Rate/Positive Predictive Value was 90.23 ± 4.29%, while Recall Rate/Sensitivity was 88.23 ± 4.84%. These factors and average counting speed of our algorithm were compared with two other open source applications (QuPath and CellProfiler) and resulted in 6-7% higher Recall Rate, while 4- to 30-fold higher processing speed. In conclusion, our image analysis algorithm can reliably detect and count nuclear signals in digital whole slides or any selected large areas i.e. hot spots, thus can support pathologists in assessing clinically important nuclear biomarkers with less intra- and interlaboratory bias inherent of empirical scoring. © 2017 International Society for Advancement of Cytometry.

Local volume fraction distributions of axons, astrocytes, and myelin in deep subcortical white matter

“…Digital pathology techniques can enhance pathologists precision in biomarker assessment and accelerate diagnosis. Hence, there is an increasing interest from this community in automated image analysis technologies that support histopathological assessment of tissue structure (see for example (Paulik et al, 2017;Xu et al, 2017;Barker et al, 2016)). …”

Machine Learning with Optical Phase Signatures for Phenotypic Profiling of Cell Lines

“…In contrast, optical phase imaging and classification of substrate‐adherent cells provides additional information about cell–cell and cell–substrate interactions. Whole‐slide digital pathology and a high‐throughput well‐plate culture imaging techniques require robust, automated methods to identify cells of clinical interest. The classification accuracy from this study of 90–100% is high, similar to other reported values using machine‐learning algorithms trained on optical phase map data .…”

Quantitative chromogenic immunohistochemical image analysis in cellprofiler software

“…Designed for biologists, CellProfiler performs multiple sample analyses simultaneously, providing a high‐throughput platform without the need to adjust the configuration or the user to possess programming skills. CellProfiler has been widely used for a variety of biological applications from the measurement of cell size and assessment of cell morphology to assays determining wound healing . Most frequently, this system has been published supporting the quantification of fluorescently stained microscopy specimens.…”