A New Multi-Atlas Based Deep Learning Segmentation Framework with Differentiable Atlas Feature Warping

Huabing Liu,Dong Nie,Jian Yang,Jinda Wang,Zhenyu Tang

DOI: https://doi.org/10.1109/jbhi.2023.3344646

IF: 7.7

2024-01-01

IEEE Journal of Biomedical and Health Informatics

Abstract:Deep learning based multi-atlas segmentation (DL-MA) has achieved the state-of-the-art performance in many medical image segmentation tasks, e.g., brain parcellation. In DL-MA methods, atlas-target correspondence is the key for accurate segmentation. In most existing DL-MA methods, such correspondence is usually established using traditional or deep learning based registration methods at image level with no further feature level adaption. This could cause possible atlas-target feature inconsistency. As a result, the information from atlases often has limited positive and even counteractive impact on the final segmentation results. To tackle this issue, in this paper, we propose a new DL-MA framework, where a novel differentiable atlas feature warping module with a new smooth regularization term is presented to establish feature level atlas-target correspondence. Comparing with the existing DL-MA methods, in our framework, atlas features containing anatomical prior knowledge are more relevant to the target image feature, leading the final segmentation results to a high accuracy level. We evaluate our framework in the context of brain parcellation using two public MR brain image datasets: LPBA40 and NIREP-NA0. The experimental results demonstrate that our framework outperforms both traditional multi-atlas segmentation (MAS) and state-of-the-art DL-MA methods with statistical significance. Further ablation studies confirm the effectiveness of the proposed differentiable atlas feature warping module.