Segment Anything for Dendrites from Electron Microscopy

Zewen Zhuo; Ilya Belevich; Ville Leinonen; Eija Jokitalo; Tarja Malm; Alejandra Sierra; Jussi Tohka

cs.CVNov 4, 2024

Segment Anything for Dendrites from Electron Microscopy

Authors: Zewen Zhuo, Ilya Belevich, Ville Leinonen, Eija Jokitalo, Tarja Malm, Alejandra Sierra, Jussi Tohka

Organizations: A.I. Virtanen Institute for Molecular Sciences University of Eastern Finland Kuopio, Finland · Electron Microscopy Unit Institute of Biotechnology University of Helsinki Helsinki, Finland · Department of Medicine Faculty of Health Sciences University of Eastern Finland Kuopio, Finland

Abstract

Segmentation of cellular structures in electron microscopy (EM) images is fundamental to analyzing the morphology of neurons and glial cells in the healthy and diseased brain tissue. Current neuronal segmentation applications are based on convolutional neural networks (CNNs) and do not effectively capture global relationships within images. Here, we present DendriteSAM, a vision foundation model based on Segment Anything, for interactive and automatic segmentation of dendrites in EM images. The model is trained on high-resolution EM data from healthy rat hippocampus and is tested on diseased rat and human data. Our evaluation results demonstrate better mask quality compared to the original and other fine-tuned models, leveraging the features learned during training. This study introduces the first implementation of vision foundation models in dendrite segmentation, paving the path for computer-assisted diagnosis of neuronal anomalies.

Figures & tables

Dataset	Pixel Size	Width $\times$ Height	Slices	Employed
A	15 $\times$ 15 $nm^{2}$	3100 $\times$ 3100	1044	Training & Evaluation
B	15 $\times$ 15 $nm^{2}$	3100 $\times$ 3100	698	Evaluation
C	10 $\times$ 10 $nm^{2}$	4096 $\times$ 4096	697	Evaluation

TABLE I: Overview of Data Utilized

Fig. 1: Concavity Distribution: The x-axis and y-axis represent the value of concavity and the corresponding percentage of masks in sampled images, respectively. The green star and red arrow in the subplot indicate the dendrite and spine structure, respectively.

Fig. 3: Impact of Image Tiling vs Resizing: On the left, segmentation quality with the model trained with the tiling method (cropping with a sliding window size 1024 and step size 512), and on the right with the model trained with the resizing method (resizing the whole image to 1024 $\times$ 1024). In the legend, the digits after ”p” and ”n” stand for the number of foreground points and background points used, respectively.

Fig. 4: Quantitative Evaluation: The blue and black legends present ViT-B and ViT-L models from SAM. ViT-B-EM-organelles [ 32 ] and ViT-L-EM-organelles [ 33 ] stand for the models from Micro_SAM trained on EM organelles images, and ViT-B-resize-EM-dendrite and ViT-L-resize-EM-dendrite are specialist models that we trained.

Scale	Quality
1	Poor Quality
2	Major Part of the Object Missing
3	Object Moderately Segmented
4	Major Part of the Object Segmented
5	Acceptable Quality

TABLE II: Qualitative Grading Criteria

Fig. 5: Qualitative Analysis: Expert graded quality of randomly sampled masks predicted by ViT-L-resize-EM-dendrite with bbox_p4_n8 prompts.

Fig. 6: Quality Comparison in User Study and Efficiency Improvements: (a) masks from fully manual annotation, (b) masks from model-assisted annotation, (c) average time spent per object under different annotation modes.

Fig. 7: Mask Similarities in Different Annotation Modes: (a) & (b) are examples from dataset A, (c) & (d) are examples from dataset B.

Fig. 8: Quantitative Evaluation of Automatic Inference

Fig. 9: Qualitative Analysis for Automatic Inference

Explore similar work

$μ$ Match: Foundation Models for Semi-supervised Learning and Domain Adaptation in EM
Jun 19, 2026Marei Freitag, Olesia Korchevaia, Luca Freckmann +2Recent Vision Foundation Models Wireless Foundation Models
Life and Medical Sciences Institute (LIMES), University of Bonn, Germany · Institute of Computer Science, Georg-August-University Göttingen, Germany
Probe-EM: Targeted Neuron Tracing via Training-Free Semantic Verification
Jul 6, 2026Liuyun Jiang, Yanchao Zhang, Jinyue Guo +5Human Connectome Project Neural Reconstruction
State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences, Beijing, China · School of Future Technology, University of Chinese Academy of Sciences · School of Artificial Intelligence, University of Chinese Academy of Sciences +1
NeuroAdaptTrainer: A Fiji/ImageJ Plugin for YOLO-Based Neuron Segmentation, InteractiveCorrection and Transfer Learning
Aug 5, 2026Daniela Eraso-Casas, Gerard Villarroya-Pique, Esther Serrano-Pertierra +4Microscopy Yolo26
aComputer Sciences Department, University of Oviedo, Asturias, Spain · bElectrical Engineering Department, University of Oviedo, Asturias, Spain · cBiomedical Engineering Center (BME), University of Oviedo, Asturias, Spain +3

Jun 19, 2026, cs.CVJ/K move · Enter open · S save

$μ$ Match: Foundation Models for Semi-supervised Learning and Domain Adaptation in EM

Marei Freitag, Olesia Korchevaia, Luca Freckmann, Anwai Archit+1

Life and Medical Sciences Institute (LIMES), University of Bonn, Germany · Institute of Computer Science, Georg-August-University Göttingen, Germany

Recent Vision Foundation Models Wireless Foundation Models Domain Adaptation Electron Microscopy Semi-Supervised Learning Foundation Model+1

Open paper AskPDF arXiv

Vision foundation models have substantially advanced computer vision, enabling state-of-the-art performance in zero- and few-shot settings. They have been successfully applied to biomedical imaging tasks ranging from organ segmentation in computed tomography to cell segmentation in light microscopy. Electron microscopy (EM) is a central modality for analyzing cellular ultrastructure due to its nanometer-scale resolution. However, the application of foundation models in EM has so far been limited to specific organelles, such as mitochondria, largely due to the diversity of segmentation tasks and the scarcity of comprehensively annotated data. As a result, EM segmentation still predominantly relies on supervised learning, requiring extensive manual annotation and limiting ultrastructural analysis. To address this gap, we propose

μ

Match, a framework for semi-supervised learning and domain adaptation that leverages foundation models. We implement state-of-the-art student-teacher-based methods and evaluate multiple foundation models (SAM, SAM2,

μ

SAM, DINOv2/v3) on challenging EM tasks, including mitochondrion, nucleus, and neurite segmentation. Our results demonstrate consistent improvements over strong baselines and highlight a path toward substantially reducing the annotation effort in EM.

Abstract

Explore similar work

μμμMatch: Foundation Models for Semi-supervised Learning and Domain Adaptation in EM

Probe-EM: Targeted Neuron Tracing via Training-Free Semantic Verification

NeuroAdaptTrainer: A Fiji/ImageJ Plugin for YOLO-Based Neuron Segmentation, InteractiveCorrection and Transfer Learning

$μ$ Match: Foundation Models for Semi-supervised Learning and Domain Adaptation in EM