Unsupervised Domain Adaptation for Vestibular Schwannoma and Cochlea Segmentation via Semi-supervised Learning and Label Fusion

Automatic methods to segment the vestibular schwannoma (VS) tumors and cochlea from magnetic resonance imaging (MRI) are critical VS treatment planning. Although supervised have achieved satisfactory performance in segmentation, they require full annotations by experts, which is laborious time-consuming. In this work, we aim tackle segmentation problem an unsupervised domain adaptation setting. Our proposed method leverages both image-level alignment minimize divergence semi-supervised training further boost performance. Furthermore, propose fuse labels predicted multiple models via noisy label correction. MICCAI 2021 crossMoDA challenge ( https://crossmoda.grand-challenge.org/ ), our results on final evaluation leaderboard showed that has promising with mean dice score of 79.9% 82.5% ASSD 1.29 mm 0.18 for tumor cochlea, respectively. The outperformed all other competing as well nnU-Net.