We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

q-bio.QM

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Biology > Quantitative Methods

Title: Cellcounter: a deep learning framework for high-fidelity spatial localization of neurons

Abstract: Many neuroscientific applications require robust and accurate localization of neurons. It is still an unsolved problem because of the enormous variation in intensity, texture, spatial overlap, morphology and background artifacts. In addition, curation of a large dataset containing complete manual annotation of neurons from high-resolution images to train a classifier requires significant time and effort. We present Cellcounter, a deep learning-based model trained on images containing incompletely-annotated neurons with highly-varied morphology and control images containing artifacts and background structures. Leveraging the striking self-learning ability, Cellcounter gradually labels neurons, obviating the need for time-intensive complete annotation. Cellcounter shows its efficacy over the state of the arts in the accurate localization of neurons while significantly reducing false-positive detection in several protocols.
Comments: Submitted to a journal
Subjects: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG)
Cite as: arXiv:2103.10462 [q-bio.QM]
  (or arXiv:2103.10462v1 [q-bio.QM] for this version)

Submission history

From: Tamal Batabyal [view email]
[v1] Thu, 18 Mar 2021 18:19:17 GMT (38634kb,D)

Link back to: arXiv, form interface, contact.