Publikation
Bounding Box is all you need: Learning to Segment Cells in 2D Microscopic Images via Box Annotations
Nabeel Khalid; Maria Caroprese; Gillian Lovell; Daniel A Porto; Johan Trygg; Andreas Dengel; Sheraz Ahmed
In: 28th UK Conference on Medical Image Understanding and Analysis - MIUA. Medical Image Understanding and Analysis (MIUA-2024), Spriner Nature, 8/2024.
Zusammenfassung
Microscopic imaging plays a pivotal role in various fields of
science and medicine, offering invaluable insights into the intricate world
of cellular biology. At the heart of this endeavor lies the need for accurate
identification and characterization of individual cells within these
images. Deep learning-based cell segmentation, which involves delineating
cells from complex microscopic images, is pivotal for cell analysis.
It serves as the foundation for extracting meaningful information about
cell morphology, spatial organization, and interactions. However, traditional
deep-learning models for cell segmentation require extensive and
expensive annotation masks for each cell in the image, posing a significant
challenge. To address this issue, this study introduces CellBoxify, a novel
pipeline that streamlines cell instance segmentation. Unlike traditional
methods, CellBoxify operates solely on bounding box annotations, making
it approximately seven times faster than manual segmentation mask
annotation for each cell. The proposed approach’s effectiveness is evident
in its performance on the LIVECell dataset, a well-known resource
for cell segmentation research. Achieving 83.40% of the fully supervised
performance on this dataset demonstrates the efficacy of the proposed
method.