Nirmal Das presents Multiscale segmentation of overlapping structures and its bio-medical applications

Multiscale segmentation of overlapping structures is a challenging task for the
image processing and computer vision community. Two objects, both in 2D and 3D,
may overlap in isointensity space or they may overlap in a shared intensity
space. Moreover, the objects may overlap at multiple places with varying
degrees
of overlap, making the separation process more complex. Segmentation problems
like segmentation carotid vasculature from brain CT images, separation of
pulmonary artery-vein from lung CT images, segmentation of dendritic spines
from
dendrite from 3D microscopic images etc. can be formulated as multiscale
segmentation of overlapping structures. In this talk, the multiscale
segmentation problem will be discussed in the context of 1) 3D segmentation of
the dendritic spine from microscopic images and 2) segmentation of overlapping
cell nuclei from 2D images:

3dSpAn: a tool for segmentation and quantitative morphological analysis of
individual dendritic spines in 3D using the Multiscale Opening (MSO) algorithm
https://link.springer.com/article/10.1007/s12021-021-09549-0

Deep-Fuzz: integration of deep learning and fuzzy waterflow for segmentation of
overlapping cell nuclei.
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0286862

[Nirmal Das has just started to work as a post-doc in the group of Jan Kybic.]