Inter-subject registration-based one-shot segmentation with alternating union network for cardiac MRI images

• Published in: Medical image analysis Vol. 79; p. 102455
• Main Authors: Wang, Heying, Li, Qince, Yuan, Yongfeng, Zhang, Ze, Wang, Kuanquan, Zhang, Henggui
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.07.2022; Elsevier BV
• Subjects: Cardiac images; Deep learning; Heart; Image processing; Image segmentation; Labels; Magnetic resonance imaging; Medical image segmentation; Medical imaging; Registration; Similarity; Deep learning; Cardiac images; Registration; Medical image segmentation
• ISSN: 1361-8415; 1361-8423; 1361-8423
• DOI: 10.1016/j.media.2022.102455

•A novel way of one-shot segmentation is provided. Medical images can be directly segmented based on inter-subject registration, using only one labeled volume and a series of unlabeled scans. The proposed framework is extensible to incorporate extra labeled volumes.•Intensity independent LSNs are innovatively introduced in AUN. They guide AUN to optimize the segmentation results by paying more attention to intrinsic anatomical characteristics, eliminating negative optimization caused by the complex background.•A new similarity measurement named LSE is proposed in this study. Combining the reconstructed image using medians, it can significantly enhance the frameworks ability to segment images with complex intensity.•The proposed framework outperforms state-of-the-art baseline methods in terms of performance, laying a foundation for further research on one-shot segmentation. [Display omitted] Medical image segmentation based on deep-learning networks makes great progress in assisting disease diagnosis. However, currently, the training of most networks still requires a large amount of data with labels. In reality, labeling a considerable number of medical images is challenging and time-consuming. In order to tackle this challenge, a new one-shot segmentation framework for cardiac MRI images based on an inter-subject registration model called Alternating Union Network (AUN) is proposed in this study. The label of the source image is warped with deformation fields discovered from AUN to segment target images directly. Initially, the volumes are pre-processed by aligning affinely and adjusting the global intensity to simplify subsequent deformation registration. AUN consists of two kinds of subnetworks trained alternately to optimize segmentation gradually. The first kind of subnetwork takes a pair of volumes as inputs and registers them using global intensity similarity. The second kind of subnetwork, which takes the predicted labels generated from the previous subnetwork and the labels refined using the information of intrinsic anatomical structures of interest as inputs, is intensity-independent and focuses attention on registering structures of interest. Specifically, the input of AUN is a pair of a labeled image with the texture in regions of interest removed and a target image. Additionally, a new similarity measurement more appropriate for registering such image pair is defined as Local Squared Error (LSE). The proposed registration-based one-shot segmentation pays attention to the problem of the lack of labeled medical images. In AUN, only one labeled volume is required and a large number of unlabeled ones can be leveraged to improve segmentation performance, which has great advantages in clinical application. In addition, the intensity-independent subnetwork and LSE proposed in this study empower the framework to segment medical images with complicated intensity distribution.