Soma instance segmentation, i.e., identifying and delineating individual cell somas as distinct instances, is crucial for cellular analysis and connectomic reconstruction. Three-dimensional electron microscopy (3D EM) provides nanometer-scale resolution for capturing fine-grained soma morphology. However, dense instance-level manual annotation is prohibitively costly, limiting the scalability of fully supervised methods. To address this challenge, we propose SomaNet, a weakly supervised framework for 3D EM soma instance segmentation under partial annotation constraints. SomaNet adopts a teacher--student learning paradigm tailored to partial labels. The teacher is trained using partially annotated data to generate pseudo-labels, while the student jointly learns from the partial ground-truth annotations and the generated pseudo-labels, progressively recovering dense instance segmentations. To accurately delineate cell somas under limited supervision with varying instance counts, SomaNet incorporates affinity learning, which encourages high similarity within instances and low similarity across instance boundaries. Semantic-guided affinity decoding and 2D-to-3D reconstruction then produce volumetrically consistent 3D soma instances while preserving the large receptive fields of 2D backbones. The framework is architecture-flexible and supports diverse backbones, including vision transformers and foundation models, enabling direct transfer of pretrained visual representations to volumetric EM segmentation. Experiments on 3D EM brain datasets demonstrate that SomaNet achieves accurate and robust soma instance segmentation across regions with diverse soma morphologies under partial annotation. Code is available at https://github.com/mkhateri/SomaNet.
Figures & tables
Fig. 1: Overview of the proposed teacher–student framework. The teacher (top) processes the original image to generate pseudo-labels y~j=\mathbbm1[p^t(1)>τ] . The student (bottom) receives K augmented views and is trained with supervised and consistency objectives. Supervision on labeled regions is applied via Lsup , combining mask and affinity losses. Consistency is enforced through pseudo-labels via Lmasku , while feature-level alignment between teacher and student is imposed by Ldistill . The teacher parameters are updated as an exponential moving average of the student parameters. The affinity visualization shows the mean over the nearest-neighbor offset channels in R , where brighter regions indicate higher predicted similarity between adjacent pixels and darker regions correspond to instance boundaries.
Fig. 2: Propagation of partial annotations into dense affinity during SomaNet training. (a) Raw EM input, (b) dense affinity predicted by SomaNet, and (c) partially annotated ground truth covering only a subset of soma instances. The proposed framework enables dense prediction from limited annotations.
Fig. 3: Backbone adaptation in SomaNet. Existing encoder–decoder architectures are adapted with minimal modifications. The decoder produces a shared embedding representation that is used for affinity learning to learn discriminative voxel features for instance segmentation, while the same decoder features are simultaneously provided to a semantic segmentation head for foreground mask prediction.
Dataset
Method
Dice ↑
IoU ↑
PQ ↑
ARAND ↓
VOI total↓
VOI split↓
VOI merge↓
Test Block 1
EMADS (3D U-Net)
0.835
0.717
0.643
0.737
2.417
0.853
1.564
EMADS (Swin UNETR)
0.742
0.590
0.625
0.822
2.723
0.657
2.066
SomaNet (DINOv2) [Ours]
0.993
0.986
0.706
0.104
0.411
0.244
0.166
SomaNet (SwinIR) [Ours]
0.991
0.982
0.767
0.103
0.454
0.220
0.233
Test Block 2
EMADS (3D U-Net)
0.793
0.657
0.563
0.780
2.653
0.841
1.812
EMADS (Swin UNETR)
0.810
0.681
0.692
0.729
2.329
0.708
1.621
TABLE I: Quantitative comparison of soma instance segmentation methods on 3D EM test sets. Best results are highlighted in bold .
Fig. 4: Qualitative comparison of the proposed SomaNet and the EMADS baselines on the two test sets. Columns show (a) input EM image, (b) EMADS (Swin UNETR), (c) EMADS (3D U-Net), (d) SomaNet (DINOv2), (e) SomaNet (SwinIR), and (f) ground-truth annotation. Across both test sets, SomaNet produces more accurate, boundary-preserving, and topologically consistent soma instance segmentations, particularly in densely packed regions. Nevertheless, it occasionally misses very small soma cells (white arrow) in Test Block 2.
Fig. 5: Qualitative results on four representative training volumes selected from the partially annotated EMADS training dataset. From left to right: (a) input 3D EM volume, (b) dense 3D soma instance segmentation generated by SomaNet, (c) representative EM slice, (d) predicted instance labels on the corresponding slice, and (e) available partial ground-truth annotations used for supervision. Despite supervision from only a subset of annotated soma instances, SomaNet successfully propagates the available annotations to recover dense, boundary-accurate, and topologically consistent 3D soma instance segmentations across volumes exhibiting diverse soma distributions, morphologies, and imaging characteristics.
Fig. 6: Robustness of SomaNet under challenging ssTEM imaging conditions. (a) Representative EM image with a characteristic line artifact and low-contrast soma boundaries. (b) Corresponding SomaNet prediction, accurately delineating individual soma instances while preserving clear separation between adjacent cells.
Fig. 7: Representative failure case of SomaNet. The white arrow indicates a false-positive soma detection.
Accurate three-dimensional (3D) reconstruction of individual dendrites in serial block-face scanning electron microscopy (SBF-SEM) is essential for quantifying structural plasticity in the brain, yet manual annotation at scale is infeasible. We present a fully automatic pipeline for 3D dendrite instance segmentation that unifies YOLOv6-guided Segment Anything Model (SAM) prompting on downsampled slices, iterative two-dimensional mask refinement, random forest 3D instance linking, and instance-aware high-resolution refinement using nnU-Net at native resolution into a single system requiring no manual prompting at inference. Applied to hippocampal CA1 SBF-SEM datasets from a control rat and a pilocarpine- induced epileptic rat, our pipeline reconstructs coherent, well- separated dendrites with high semantic accuracy (Dice 0.93 and 0.91) and strong instance-level performance on control tissue, while analysis of the more challenging epileptic tissue identifies instance recognition in dense regions as the principal remaining limitation. The high-resolution refinement stage recovers thin dendritic protrusions, providing a basis for downstream spine- level analysis. Code is available at https://github.com/ ZE-WEN/dendrite-3d-instance-seg.
Zewen Zhuo, Ilya Belevich, Eija Jokitalo +2
A.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland
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 μMatch, 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, μSAM, 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.
Accurate cell segmentation in pathology images typically requires dense pixel-wise annotations, which are costly and time-consuming to obtain. This challenge is especially important for emerging biological imaging modalities and multiplexed datasets with variable channel configurations, where expert-labeled data are scarce. In this work, we introduce ImPartial, a deep learning framework designed to achieve state-of-the-art segmentation performance in low-annotation regimes using sparse scribbles and limited supervision. ImPartial augments the segmentation objective via self-supervised multi-channel quantized imputation. This approach leverages the observation that perfect pixel-wise reconstruction or denoising of the image is not needed for accurate segmentation, and thus, introduces a self-supervised classification objective that better aligns with the overall segmentation goal. We demonstrate that ImPartial achieves performance at par with fully supervised models while requiring substantially fewer annotations. Extensive experiments on benchmark multiplexed cellular imaging and single-plex clinical brightfield immunohistochemistry datasets show consistent improvements over strong baselines with only partial annotations. All benchmark datasets and code are available via our Github: https://github.com/nadeemlab/ImPartial.
Gunjan Shrivastava, Saad Nadeem
Memorial Sloan Kettering Cancer Center, New York NY 10065, USA