Multi-task Attention-Based Semi-supervised Learning for Medical Image Segmentation

TY - GEN

T1 - Multi-task Attention-Based Semi-supervised Learning for Medical Image Segmentation

AU - Chen, Shuai

AU - Bortsova, Gerda

AU - Juárez, Antonio García Uceda

AU - van Tulder, Gijs

AU - de Bruijne, Marleen

PY - 2019/1/1

Y1 - 2019/1/1

N2 - We propose a novel semi-supervised image segmentation method that simultaneously optimizes a supervised segmentation and an unsupervised reconstruction objectives. The reconstruction objective uses an attention mechanism that separates the reconstruction of image areas corresponding to different classes. The proposed approach was evaluated on two applications: brain tumor and white matter hyperintensities segmentation. Our method, trained on unlabeled and a small number of labeled images, outperformed supervised CNNs trained with the same number of images and CNNs pre-trained on unlabeled data. In ablation experiments, we observed that the proposed attention mechanism substantially improves segmentation performance. We explore two multi-task training strategies: joint training and alternating training. Alternating training requires fewer hyperparameters and achieves a better, more stable performance than joint training. Finally, we analyze the features learned by different methods and find that the attention mechanism helps to learn more discriminative features in the deeper layers of encoders.

AB - We propose a novel semi-supervised image segmentation method that simultaneously optimizes a supervised segmentation and an unsupervised reconstruction objectives. The reconstruction objective uses an attention mechanism that separates the reconstruction of image areas corresponding to different classes. The proposed approach was evaluated on two applications: brain tumor and white matter hyperintensities segmentation. Our method, trained on unlabeled and a small number of labeled images, outperformed supervised CNNs trained with the same number of images and CNNs pre-trained on unlabeled data. In ablation experiments, we observed that the proposed attention mechanism substantially improves segmentation performance. We explore two multi-task training strategies: joint training and alternating training. Alternating training requires fewer hyperparameters and achieves a better, more stable performance than joint training. Finally, we analyze the features learned by different methods and find that the attention mechanism helps to learn more discriminative features in the deeper layers of encoders.

KW - Attention

KW - Brain tumor

KW - Deep learning

KW - Multi-task learning

KW - Segmentation

KW - Semi-supervised learning

KW - White matter hyperintensities

UR - http://www.scopus.com/inward/record.url?scp=85075682442&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-32248-9_51

DO - 10.1007/978-3-030-32248-9_51

M3 - Article in proceedings

AN - SCOPUS:85075682442

SN - 9783030322472

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 457

EP - 465

BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings

A2 - Shen, Dinggang

A2 - Yap, Pew-Thian

A2 - Liu, Tianming

A2 - Peters, Terry M.

A2 - Khan, Ali

A2 - Staib, Lawrence H.

A2 - Essert, Caroline

A2 - Zhou, Sean

PB - Springer VS

T2 - 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2019

Y2 - 13 October 2019 through 17 October 2019

ER -