importance 4/10partially solvedDomain Adaptation and Few-Shot Learningneuroscience imagingstage: researchopen-source model
Problem
Deep learning models for mitochondria segmentation in electron microscopy volumes need large annotated datasets and perform poorly on EM data from different tissues or species (domain shift).
Why it matters
Accurate organelle-level segmentation is essential to characterize neuronal structures, and annotating large EM datasets is costly.
What the AI did
The paper presents three unsupervised deep-learning domain adaptation strategies to improve mitochondria segmentation on target-domain EM datasets.
Results
not stated in source
Impact
Proposed strategies to improve cross-dataset mitochondria segmentation; broader impact not stated in source.
Record
Venue
Computer Methods and Programs in Biomedicine
Published
2022
Work type
article
Authors
Daniel Franco-Barranco, Julio Pastor-Tronch, Aitor González-Marfil, Arrate Muñoz‐Barrutia, Ignacio Arganda‐Carreras
deep learning, domain adaptation, mitochondria segmentation, electron microscopy, unsupervised
Importance rating
4/ 10
real but narrower advance
Domain adaptation for mitochondria segmentation across tissues/species reduces annotation burden for EM brain volumes; real but narrow organelle-segmentation advance.
Basis: inferred.
Judged from the source text, which gives no explicit scale.
Stated scale: not stated in source. Method and limits.