Organizations: School of Engineering and Physical Science, University of Lincoln, Lincoln, UK · School of Computer Science, University of Nottingham, Nottingham, UK · Queen’s Medical Centre, University of Nottingham, Nottingham, UK · Department of Data Science and Artificial Intelligence, The Hong Kong Polytechnic University, Hong Kong SAR
Multi-modal brain tumor segmentation is essential for clinical diagnosis, radiotherapy planning, and longitudinal monitoring. While recent studies have achieved remarkable overall segmentation performance, they rely on overlap-based objectives that inherently favor large regions, causing sub-3 mm micro-lesions to be systematically suppressed during training. Moreover, existing pixel-level segmentation frameworks lack explicit instance-level modeling, leading to unreliable lesion counting that directly impacts clinical staging and therapeutic decisions. To address these challenges, we propose MicroBT, a unified multi-modal framework for micro brain tumor segmentation and counting. Our core insight is that a fundamental geometric property is inherently scale-invariant: regardless of size, each lesion possesses a well-defined centroid. Leveraging this, we first devise a centroid-anchored synergistic paradigm that utilizes predicted centroids as shared anchors to simultaneously condition dense segmentation and instance-level localization. To counteract the tumor size imbalance, we further devise a Volume-Calibrated Loss (VC-Loss) that adaptively amplifies the gradient contribution of micro-lesions, alongside a Size-Aware Curriculum Optimization (SACO) that stabilizes multi-task training by establishing global representations before progressively unlocking fine-grained localization. Extensive experiments demonstrate that MicroBT outperforms state-of-the-art methods, particularly on metastases, where micro-lesion (≤3 mm) localization precision improves by approximately 7% and counting estimation errors decrease by approximately 26%. The code is available at https://github.com/leonlou921/MicroBT.
Figures & tables
Figure 1: Comparison of segmentation quality and counting reliability across four BraTS datasets. Our MicroBT framework achieves the best segmentation accuracy and counting performance with comparable computational cost to existing methods, demonstrating that centroid-guided synergistic learning effectively bridges dense segmentation and instance-level localization for multi-modal micro brain tumors.
Figure 2: The overview of our MicroBT framework for multi-modal micro brain tumor segmentation and counting. MicroBT utilizes predicted centroids as shared anchors to simultaneously condition dense segmentation and instance-level localization.
Method
Metastases
Meningiomas
Gliomas
Average
Dice ↑
HD95 ↓
Dice ↑
HD95 ↓
Dice ↑
HD95 ↓
Dice ↑
HD95 ↓
nnU-Net [ 15 ]
72.15
24.65
80.85
19.62
88.52
6.75
80.51
17.01
TransUNet [ 7 ]
65.76
28.16
79.11
26.62
88.28
8.76
77.72
21.18
nnFormer [ 48 ]
67.39
26.10
76.36
21.82
86.91
9.02
76.89
18.98
Swin-UMamba [ 25 ]
68.85
19.45
79.82
15.85
88.35
6.65
79.01
13.98
UMamba [ 30 ]
64.07
29.93
78.14
30.83
87.55
10.55
76.59
23.77
Table 1: Comparison with state-of-the-arts on Micro Brain Tumor Segmentation.
Method
Metastases
Meningiomas
Gliomas
Average
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
nnDetection [ 3 ]
0.852
25.58
1.865
38.55
1.017
39.53
1.245
34.55
nnU-Net [ 15 ]
0.568
47.51
0.536
64.18
0.598
65.49
0.567
59.06
TransUNet [ 7 ]
0.907
30.46
1.894
33.94
1.011
47.66
1.271
37.35
nnFormer [ 48 ]
0.701
48.66
0.992
42.64
1.015
46.01
0.903
45.77
Swin-UMamba [ 25 ]
0.773
28.36
1.812
41.33
1.019
38.04
1.201
35.91
Table 2: Comparison with state-of-the-arts on Micro Brain Tumor Counting.
Figure 3: Statistical significance analysis via paired t -tests across all test cases. Top: MicroBT vs. nnU-Net [ 15 ] on segmentation metrics (Dice, HD95). Bottom: MicroBT vs. nnMamba [ 13 ] on counting metrics (MAPE, AP 50 ). **: p<0.01 , *: p<0.05 , ns: not significant. These results demonstrate that our MicroBT framework achieves significant improvements on nearly all metrics across three tumor types.
Figure 4: Qualitative comparison of segmentation and counting results across different methods. Benefiting from the centroid-anchored synergistic paradigm, the proposed MicroBT framework demonstrates superior micro-lesion sensitivity and more accurate lesion counting compared to state-of-the-art methods.
CAS
VC-Loss
SACO
Metastases
Meningiomas
Gliomas
Average
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
MAPE ↓
AP 50 ↑
5.451
53.39
0.749
53.52
0.889
70.27
2.363
59.06
✓
1.385
60.00
0.454
66.20
0.678
71.81
0.839
66.00
✓
✓
1.082
62.63
0.420
67.06
0.574
77.93
0.692
69.21
✓
✓
✓
0.423
65.96
0.401
68.52
0.565
79.56
0.463
71.35
Table 3: Ablation study of our MicroBT framework on Micro Brain Tumor Counting.
Figure 5: Hyperparameter sensitivity analysis on Top- K sampling number, number of object queries q , and VC-Loss scaling factor α . All three hyperparameters exhibit consistent trends between Dice and AP 50 , with optimal performance achieved at K =50, q =50, and α =0.5.
Figure 6: Object-level FROC curves across three size regimes. MicroBT achieves the highest sensitivity at all false-positive rates. On the ≤ 3 mm subset, MicroBT maintains ∼ 68% sensitivity while competitors plateau below 55% ∼ 70%.
Figure 7: Case-level confusion matrices of nnMamba [ 13 ] and MicroBT across three size regimes. Note that case counts vary as each threshold yields a different subset of lesion-bearing cases. MicroBT reduces missed cases from 117 to 94 on ≤ 3mm lesions and achieves near-perfect case-level sensitivity on all tumors with only 1 missed case.
Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.
Faculty of Information Technology, Van Lang School of Technology, Van Lang University, Ho Chi Minh City 70000, Vietnam · Institute of Information Technology and Electrical-Electronics Engineering, Ho Chi Minh City University of Transport, Ho Chi Minh City 70000, Vietnam