Organizations: Institute of Medical Technology, Peking University Health Science Center, Beijing, China · Beijing Key Laboratory of Intelligent Neuromodulation and Brain Disorder Treatment, Beijing, China · Department of Radiation Oncology, University Hospital Erlangen, Erlangen, Germany
Background: Deep learning has demonstrated significant potential for automated brain metastases (BM) segmentation; however, models trained at a singular institution often exhibit suboptimal performance at various sites due to disparities in scanner hardware, imaging protocols, and patient demographics. The goal of this work is to create a domain adaptation framework that will allow for BM segmentation to be used across multiple institutions. Methods: We propose a VAE-MMD preprocessing pipeline that combines variational autoencoders (VAE) with maximum mean discrepancy (MMD) loss, incorporating skip connections and self-attention mechanisms alongside nnU-Net segmentation. The method was tested on 740 patients from four public databases: Stanford, UCSF, UCLM, and PKG, evaluated by domain classifier's accuracy, sensitivity, precision, F1/F2 scores, surface Dice (sDice), and 95th percentile Hausdorff distance (HD95). Results: VAE-MMD reduced domain classifier accuracy from 0.91 to 0.50, indicating successful feature alignment across institutions. Reconstructed volumes attained a PSNR greater than 36 dB, maintaining anatomical accuracy. The combined method raised the mean F1 by 11.1% (0.700 to 0.778), the mean sDice by 7.93% (0.7121 to 0.7686), and reduced the mean HD95 by 65.5% (11.33 to 3.91 mm) across all four centers compared to the baseline nnU-Net. Conclusions: VAE-MMD effectively diminishes cross-institutional data heterogeneity and enhances BM segmentation generalization across volumetric, detection, and boundary-level metrics without necessitating target-domain labels, thereby overcoming a significant obstacle to the clinical implementation of AI-assisted segmentation.
Figures & tables
Figure 1: Representative axial T1-weighted post-contrast MRI slices from four institutions demonstrating heterogeneity in lesion characteristics. Green contours indicate expert-delineated metastases. (a) Stanford: multiple small scattered metastases characteristic of miliary disease pattern. (b) UCSF: mixed lesion sizes with prominent vascular enhancement. (c) UCLM: solitary well-circumscribed metastasis. (d) PKG: large solitary tumor from a patient with primary lung cancer. These differences in lesion burden and tumor morphology contribute to the domain gap that challenges cross-institutional model generalization.
Figure 2: VAE-MMD architecture and preprocessing pipeline. (a) The encoder progressively downsamples input volumes through four convolutional blocks (1 → 32 → 64 → 128 → 256 channels) with residual connections and self-attention modules. The latent space (512 dimensions) is regularized by KL divergence and aligned across domains via MMD loss. The decoder reconstructs images using transposed convolutions with skip connections from corresponding encoder blocks. (b) VAE-MMD reconstruction for a Stanford case (stanford_Mets_010): original images (top row) are passed through the VAE encoder-decoder to produce reconstructed volumes (middle row) in which institution-specific intensity distributions and scanner-related stylistic variations are normalized while lesion contrast and anatomical structures are preserved (PSNR = 38.55 dB, MSE = 0.000140). The near-black difference maps (bottom row, scale 0–0.5) confirm minimal information loss. These domain-harmonized reconstructions serve directly as nnU-Net inputs, reducing cross-institutional domain shift without requiring any target-domain labels.
Figure 3: Domain adaptation evaluation. (a) t-SNE visualization: two-center (top) and four-center (bottom) configurations before and after VAE-MMD, showing progressive inter-institutional mixing. (b) Confusion matrices before and after VAE-MMD: accuracy declines from 91.0% to 50.0%, confirming elimination of institution-specific feature signatures.
Baseline
VAE-MMD
Configuration
Test
Sens
Prec
F1
F2
sDice
HD95
Sens
Prec
F1
F2
sDice
HD95
\multirow 2*Two-center
Stanford
0.384
0.889
0.536
0.433
0.6831
12.78
0.652
0.683
0.638
0.649
0.7827
6.76
UCSF
0.756
0.894
0.819
0.780
0.8102
2.23
0.750
0.840
0.772
0.757
0.8251
2.07
\multirow 3*Three-center
Stanford
0.455
0.446
0.412
0.424
0.6738
14.79
0.628
0.684
0.635
0.626
0.6989
10.97
UCSF
0.767
0.866
0.798
0.776
0.8109
2.43
0.786
0.894
0.826
0.799
0.8200
2.44
UCLM
0.730
0.793
0.715
0.702
0.7358
4.43
0.850
0.791
0.792
0.808
0.8144
1.37
Table 1: Comparison of segmentation performance between baseline mixed training and VAE-MMD preprocessing across multi-center configurations. Sens: sensitivity; Prec: precision; sDice: surface Dice; HD95: 95th percentile Hausdorff distance (mm).
Figure 4: Comparison of the performance of four-center segmentation prior to and following VAE-MMD preprocessing. The blue bars show performance after VAE-MMD, while the orange bars show baseline performance. VAE-MMD consistently raised sensitivity, F1, and surface Dice (sDice) scores at most institutions. Stanford, UCSF, and UCLM saw the biggest drops in HD95, which means that domain adaptation made it easier to draw boundaries.
Figure 5: Qualitative comparison of segmentation results before (left, Baseline) and after (right, VAE-MMD) domain adaptation across four datasets (Stanford, UCSF, UCLM, PKG from top to bottom). Green contours indicate ground truth annotations; red contours indicate model predictions. Please zoom in the digital manuscript for better visualization.
Brain metastases are the most common intracranial malignancy, occurring in roughly 30% of patients with primary solid tumors and carrying a median survival near 5.9 months. Automated segmentation is critical for treatment planning and volumetric monitoring, but metastases are frequently small, numerous, and heterogeneous in size within a single patient. We compare a plain nnU-Net baseline, a Residual Encoder Large (ResEncL) variant, region-based training, and a Primus transformer model for BraTS-METS 2026 Task 1, using patient-grouped cross-validation to prevent leakage from the longitudinal UCSD subset. Primus (label-based) is our strongest individual model by aggregate DSC/NSD, achieving 0.710/0.761 (ET), 0.742/0.785 (TC), 0.683/0.689 (WT), and 0.531/0.436 (RC). ResEncL trails Primus on aggregate DSC/NSD but achieves substantially higher lesion-wise F1 (e.g. ET: 0.452 vs. 0.052); a probability-averaging ensemble of the two only partially preserves ResEncL's F1 advantage (ET lesion-wise F1: 0.064). We further report three postprocessing and label-reconstruction pitfalls we believe generalize beyond this challenge. Code is available at https://github.com/mahdiislam79/BraTS_METS_2026.
Mahdi Islam, Musarrat Tabassum
Center for Computational & Data Sciences, Independent University, Bangladesh · Department of Electrical and Electronic Engineering, Independent University, Bangladesh · Independent Researcher
Automated brain tumor segmentation supports diagnosis, treatment planning, and monitoring of disease progression, but building models that generalize across heterogeneous tumors and limited annotated data remains difficult. We present VGG16-MCA UNet, a hybrid architecture pairing an ImageNet-pretrained VGG16 encoder with a decoder in which a Multi-Channel Attention (MCA) module recalibrates features after each skip-connection fusion, trained with the Focal Tversky loss to counter severe foreground-background imbalance. We evaluate the model as a 2D, FLAIR-only, whole-tumor segmenter on tumor-positive slices from two public datasets: the BraTS 2020 benchmark and the LGG MRI Segmentation dataset. Using 5-fold cross-validation and a single network formed by averaging the weights of the five fold models, the method attains an aggregate pixel-level Dice (F1) of 95.10% on our held-out BraTS 2020 split and 88.32% on LGG. These scores are computed over all test pixels pooled into a single confusion matrix rather than averaged per case, and are therefore not directly comparable to the per-case mean Dice used in the BraTS challenge protocol. All partitions were drawn over individual slices rather than over patients, so every patient contributes slices to both training and test; the figures above therefore measure interpolation within known patients and should be read as an upper bound rather than as generalization to new ones. The model segments a 256x256 slice in 66.32 ms on a single 6 GB NVIDIA RTX 2060, approximately 8 ms more than an equivalent VGG16-UNet without MCA. We release the split records and report the protocol in full, with the aim of providing a precisely specified and reproducible 2D FLAIR baseline.
Shubham Gajjar, Deep Joshi, Avi Poptani +1
Northeastern University, Portland, ME, USA · LDRP Institute of Technology and Research, Kadi Sarva Vishwavidyalaya, Sarva Vidyalaya Kelavani Mandal, Gandhinagar, Gujarat, INDIA
Brain metastases exhibit high inter-lesion variability in size, enhancement pattern, and post-treatment appearance, making volumetric segmentation of both pre- and post-treatment cases the central challenge of the BraTS 2026 Task 1 (Brain Metastases). We build a pragmatic pipeline on a 5-fold nnU-Net ResEnc-L ensemble, in which each fold is trained independently for 1,000 epochs with the standard Dice + cross-entropy loss on 1,296 four-modality training cases. This ensemble is followed by a rule-based post-processing cascade tuned for the lesion-wise Dice similarity coefficient (LW-DSC), a detection-oriented metric that behaves very differently from the traditional global Dice. The final pipeline reaches an LW-DSC of 0.733 / 0.751 / 0.713 / 0.549 on the enhancing tumour (ET), tumour core (TC), whole tumour (WT), and resection cavity (RC) sub-regions on the official validation leaderboard. Rather than trusting these leaderboard gains, we audit every post-processing stage with a five-fold out-of-fold (OOF) analysis with no model-training leakage over all 1,296 training cases, scored with the official BraTS evaluation code (BraTS_evaluation): it confirms two stages as robust, per-fold-consistent improvements while the third improves only the leaderboard and does not reproduce out-of-fold. We further provide a mechanistic analysis of the LW-DSC metric that explains why recall-recovering post-processing carries low risk whereas component deletion does not, and we report thirteen negative results spanning loss engineering, alternative backbones, and inference-time settings, several of which run counter to widely held intuitions. Source code is released under Apache-2.0 at https://github.com/hornbeamliu/brats2026-met.
Haobin Liu, Xin Wang
College of Software, Jilin University, Changchun, China · Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun, China