Semi-supervised multi-organ segmentation should learn more than a dozen organs from a handful of labeled scans, and the smallest organs occupy less than one percent of the foreground. We identify a failure mode of self-training that is specific to this regime. When a teacher misses a small organ in a scan, every voxel of that organ receives the confident pseudo-label background, and the student is trained toward the organ absence. Repeated over many scans, this loop extinguishes tail organs. In a pilot study, removing the gallbladder from the pseudo-labels reduced its Dice on AMOS from 36.5 to 1.0. We argue that a missing organ should be treated as unresolved rather than absent and propose Existence-aware Self-Training (ExiST). It builds pseudo-labels by leave-one-out consensus of independently trained networks, screens them with a quality gate on validation scans, and infers, per scan and per organ, whether the organ was found. For every unresolved organ, the background voxels of that scan receive the set-valued target, trained with a marginal cross-entropy that accepts either answer and a Dice term that skips uncertain voxels. We show that this loss never pushes a candidate organ toward absence and never favors one member of the set, which leaves the decision to labeled evidence. ExiST outperforms the current state-of-the-art methods and reaches 77.1 mean Dice on Synapse with 20% labeled scans and 75.9 on AMOS with 5% labeled scans.
Figures & tables
Figure 1: Per-organ Dice (lines) on (a) Synapse (20% labeled) and (b) AMOS (5% labeled). Organs are ordered by their share of foreground voxels in the labeled scans (bars).
Figure 2: Overview of ExiST. Frozen foundation networks produce leave-one-out consensus pseudo-labels, a quality gate screens them, and the local doubt (Rule A) and missing-organ (Rule B) relaxations turn accepted background into set-valued targets for a trained student.
Method
Avg.
Avg.
Dice (%) of each class
Dice ↑
ASD ↓
Sp
RK
LK
Ga
Es
Li
St
Ao
IVC
PSV
Pa
RAG
LAG
V-Net (fully supervised) [ 37 ]
62.09 ±1.2
10.28 ±3.9
84.6
77.2
73.8
73.3
38.2
94.6
68.4
72.1
71.2
58.2
48.5
17.9
29.0
General
UA-MT [ 57 ]
20.26 ±2.2
71.67 ±7.4
48.2
31.7
22.2
0.0
0.0
81.2
29.1
23.3
27.5
0.0
0.0
0.0
0.0
URPC [ 33 ]
25.68 ±5.1
72.74 ±15.5
66.7
38.2
56.8
0.0
0.0
85.3
33.9
33.1
14.8
0.0
5.1
0.0
0.0
CPS [ 9 ]
33.55 ±3.7
41.21 ±9.1
62.8
55.2
45.4
35.9
0.0
91.1
31.3
41.9
49.2
8.8
14.5
0.0
0.0
SS-Net [ 54 ]
35.08 ±2.8
50.81 ±6.5
62.7
67.9
60.9
34.3
0.0
89.9
20.9
61.7
44.8
0.0
8.7
4.2
0.0
Table 1: Comparison with state-of-the-art methods on Synapse (20% labeled), mean ±std over three runs. Best in bold, second best underlined. † Trained at 1×1×2 mm and scored on the DHC grid. Other rows as reported by DCA [ 32 ] .
Method
Avg.
Avg.
Dice (%) of each class
Dice ↑
ASD ↓
Sp
RK
LK
Ga
Es
Li
St
Ao
IVC
Pa
RAG
LAG
Du
Bl
P/U
V-Net (fully supervised) [ 37 ]
76.50
2.01
92.2
92.2
93.3
65.5
70.3
95.3
82.4
91.4
85.0
74.9
58.6
58.1
65.6
64.4
58.3
General
UA-MT [ 57 ]
42.16
15.48
59.8
64.9
64.0
35.3
34.1
77.7
37.8
61.0
46.0
33.3
26.9
12.3
18.1
29.7
31.6
URPC [ 33 ]
44.93
27.44
67.0
64.2
67.2
36.1
0.0
83.1
45.5
67.4
54.4
46.7
0.0
29.4
35.2
44.5
33.2
CPS [ 9 ]
41.08
20.37
56.1
60.3
59.4
33.3
25.4
73.8
32.4
65.7
52.1
31.1
25.5
6.2
18.4
40.7
35.8
SS-Net [ 54 ]
33.88
54.72
65.4
68.3
69.9
37.8
0.0
75.1
33.2
68.0
56.6
33.5
0.0
0.0
0.0
0.2
0.2
Table 2: Comparison with state-of-the-art methods on AMOS (5% labeled). Best in bold, second best underlined. † Mean over three seeds (std 0.07 Dice, 0.35 ASD), trained at 1×1×2 mm and scored on the 80×160×160 DHC grid. Other rows as reported by DCA [ 32 ] .
Figure 3: Qualitative comparison on the Synapse test scan with the median Dice of ExiST (20% labeled). The slices pass through the reference pancreas, and the white boxes mark the enlarged regions.
Figure 4: t-SNE of per-voxel features before the classifier on the Synapse test set (40 voxels per organ and scan).
Variant
Dice
ASD
Tail
(a) Foundation only, no self-training
74.23 ±0.7
2.70 ±0.2
63.1
(b) (a) + consensus, no relaxation
75.20 ±0.6
2.76 ±0.0
63.9
(c) (b) + LDR only †
75.77 ±0.5
2.68 ±0.2
64.8
(d) (b) + MOR only †
72.83 ±0.9
6.71 ±0.8
66.2
(e) (b) + LDR + MOR, standard Dice †
73.53 ±1.0
3.05 ±0.7
63.8
(f) Full †
73.67 ±0.6
5.79 ±1.3
66.3
Table 3: Component ablation on Synapse (20% labeled). Tail is the mean Dice over Es, Pa, RAG, and LAG. † Earlier gate configuration, under which the gallbladder is gate-ignored.
Pseudo-labeler
Δ Dice
Useful:harm.
Ga Dice
EMA teacher, hard labels
+0.07
3.76
65.8→66.1
Consensus + LDR + MOR (ours)
n/a
2.21
65.8→71.3
Table 4: Pseudo-labeler design on AMOS. Δ Dice is teacher minus student, and Ga Dice is given before and after self-training.
Pseudo-labeling is a strong paradigm for semi-supervised medical image segmentation, yet its effectiveness is highly sensitive to confidence thresholding. In abdominal multi-organ segmentation, a fixed global threshold is particularly suboptimal because organ classes differ substantially in size, appearance, and learning difficulty. In this work, we propose ThreshGuide, a class-aware threshold adaptation framework that uses labeled data to guide pseudo-label selection on unlabeled data. Built upon a standard teacher-student architecture, the teacher model evaluates labeled samples during training to estimate class-aware threshold targets by maximizing an error-aware F\b{eta} criterion that balances precision and coverage. These targets are then smoothed with an exponential moving average (EMA) and used to filter unlabeled voxels in a class-dependent manner. Experiments on FLARE2022 and AMOS2022 show that ThreshGuide performs competitively overall, yielding clear improvements specifically on hard-to-learn organs.
Hongyu Liu, Yinlong Wang, Lusha Li +1
School of Intelligent Science and Technology, Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, Hangzhou, China
Multi-organ segmentation is often challenged by partially annotated datasets and domain shifts across different imaging sources. To address these limitations, we propose a two-stage learning framework that efficiently leverages partial supervision. In the first stage, the model learns from available annotations to produce accurate segmentations of annotated organs, establishing robust feature representations. In the second stage, we introduce learnable organ prototypes and a Sinkhorn-triplet loss to enforce organ-wise feature consistency across datasets. This encourages latent embeddings of the same organ to remain close, while increasing separation between different organs, even when annotations are missing. Our approach achieves performance comparable to state-of-the-art methods on the BTCV dataset, while remaining computationally efficient. By explicitly aligning feature distributions rather than relying solely on pseudo-labels, the framework effectively mitigates domain shift, making it particularly suitable for medical image segmentation tasks with limited annotation resources.
Multi-organ segmentation using deep learning requires large amounts of annotated patient data; however, institutions often lack sufficiently large and diverse annotated datasets. Privacy constraints further prevent institutions from sharing patient data to overcome this limitation. Moreover, due to the labor-intensive nature of annotation and the scarcity of diverse expertise, institutions typically have labels for only a small portion of their local data, leaving the larger unlabeled portion unused. In this work, we propose a flexible semi-supervised federated multi-task student-teacher framework that leverages federated learning (FL) to improve multi-organ segmentation using both labeled and unlabeled data across participating sites. At each communication round, the proposed framework initiates local training, where clients with labels for the same task form a federation to produce an aggregated teacher model. The resulting teachers generate task-specific features for all data at each client. Subsequently, all clients form a second federation to train a multi-task student model with a shared encoder and task-specific decoders that replicate the teacher-generated features across all segmentation tasks. The aggregated student model is then used to update the local teachers and initiate the next training round. Extensive experiments demonstrated the effectiveness of the proposed method compared with local and federated single-organ models, yielding an average performance gain of 13 percent across clients. The experiments also demonstrated the impact of multi-task learning and unlabeled data and the applicability of the framework in relaxing labeled-data requirements for client participation. The code is available at https://github.com/AshknMrd/FedMust.
Department of Circulation and Medical Imaging, Norwegian University of Science and Technology, Trondheim, Norway · Central Staff at St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway