is an examination of tissue under a microscope to identify the presence of any
disease, in digital pathology we use image-based
data for the same i.e. converting glass slides into digital images. In the modern era,
the nuclei detection, segmentation, and
classification of cancer diagnosis are
rapidly evolving, with the role of minimizing human intervention. Studies have
been conducted for numerous cancer detection and grading applications on brain,
breast, cervix, lung, prostate etc. This study provides an overview of
various nuclei detection, segmentation, feature extraction and classification
techniques and different problems associated with them.
Nucleus Segmentation is a very crucial step which leads up to identification
or grading of any disease an example for such case can be a cancer diagnosis.
There is an underlying theme for already existing techniques/strategies which is
of foreground from background
the remaining objects if they form any groups or clumps
Identification of the nuclei and growing markers
to its boundaries
developing different regions and then selecting the best for final processing
Definitely, there are some other techniques
which don’t tend to follow this pattern, but on a general scale, most of the methods/strategies applied til date do. With that said there exists a vast majority of segmentation techniques based
on a few very commonly used algorithms such as intensity thresholding, contours
models, morphology, watershed transform,
clustering, graph-based models and supervised classification etc.
Thresholding is the most basic method
for any image segmentation task. In this,
we replace existing pixels with either black or white depending on intensity
value of those pixels, if the intensity is less
than some selected constant then black is assigned otherwise white is assigned.
Hence we can also see that thresholding can result in a binary image. Therefore
the selection of the threshold value is very important which is actually based on
specific images. So different thresholding criteria are also developed so that
an optimal threshold may be selected viz histogram shape, clustering, entropy,
object attribute, spatial, local, computerized methods like Ostu and other sub-techniques
such as splitting the larger image into smaller ones and computing threshold
etc. However, for local thresholding
additional parameter is required.
Based on staining technique used nuclei can be brightly stained compared to the background hence a global threshold
can be selected for this entire
step of nucleus segmentation, for overlapping first the image is converted to the greyscale image then incremental thresholding is applied to separate the clustered