Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Domain Adaptation with SAM-RefiSeR for Enhanced Brain Tumor Segmentation

Dillan Imans, Phuoc-Nguyen Bui, Duc-Tai Le, Hyunseung Choo

TL;DR

This work tackles unsupervised domain adaptation for brain-tumor MRI segmentation under domain shift. It introduces SAM-RefiSeR, a two-phase framework that first aligns source and target appearances via frequency-domain adaptation and adversarial training, then refines target pseudo-labels with SAM-guided, morphology-aware gating in a teacher-student EMA loop. The approach yields consistent improvements over strong baselines on BraTS 2021, with the largest gains under severe modality gaps and a rapid return on early training cycles. The results demonstrate enhanced cross-domain robustness for clinically relevant tumor delineation without additional target annotations.

Abstract

Unsupervised Domain Adaptation with SAM-RefiSeR for Enhanced Brain Tumor Segmentation

Unsupervised Domain Adaptation with SAM-RefiSeR for Enhanced Brain Tumor Segmentation

TL;DR

This work tackles unsupervised domain adaptation for brain-tumor MRI segmentation under domain shift. It introduces SAM-RefiSeR, a two-phase framework that first aligns source and target appearances via frequency-domain adaptation and adversarial training, then refines target pseudo-labels with SAM-guided, morphology-aware gating in a teacher-student EMA loop. The approach yields consistent improvements over strong baselines on BraTS 2021, with the largest gains under severe modality gaps and a rapid return on early training cycles. The results demonstrate enhanced cross-domain robustness for clinically relevant tumor delineation without additional target annotations.

Abstract

Unsupervised Domain Adaptation with SAM-RefiSeR for Enhanced Brain Tumor Segmentation
Paper Structure (9 sections, 6 equations, 2 figures, 3 tables)

This paper contains 9 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Phase I: Low-Frequency Adaptation with Domain Adversarial Training
  4. Phase II: SAM‑Guided Pseudo‑label Refinement--Selection
  5. Experiments and Results
  6. Dataset and Implementation Details
  7. Comparison with Existing Methods
  8. Ablation Studies
  9. Conclusion

Figures (2)

  • Figure 1: Overview of the proposed 2-phase framework SAM-RefiSeR. (i)FFT, GRL, EMA, AF stand for (inverse) Fast Fourier Transform, Gradient Reversal Layer, Exponential Moving Average, and Amplitude Fused, respectively.
  • Figure 2: Qualitative comparison of methods with T1CE and T2 as source domains.