TY - GEN
T1 - An Ensemble Approach to Automatic Brain Tumor Segmentation
AU - Shi, Yaying
AU - Micklisch, Christian
AU - Mushtaq, Erum
AU - Avestimehr, Salman
AU - Yan, Yonghong
AU - Zhang, Xiaodong
N1 - Funding Information:
Acknowledgment. This work was performed under the Grant 2015254 and gift by the National Science Foundation and Konica Minolta, respectively. We also acknowledge support from the University of Texas at Anderson Cancer Center, Texas Advanced Computing Center, and Oden Institute for Computational and Engineering Sciences initiative in Oncological Data and Computational Science.
Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Medical image segmentation is the task of objective segmentation in medical field. 3D Tumor segmentation can help physicians efficiently diagnose cancer, track tumor change, and make treatment plans. With the development of machine learning (ML)/Deep Learning (DL) image segmentation methods, the performance of medical image segmentation has significantly improved especially in terms of accuracy and time efficiency. Performance of typical deep learning algorithms such as Fully Connection Networks, Unet, DeepLab varies with respect to different datasets, pre-processing and training parameter settings. In this paper, we propose a new architecture which utilizes the advantages of various models and aggregates their results. The original concept was inspired by Ensembles of Multiple Models and Architectures. In this paper, we train different sub-models separately. Then we train a gating network to credit the inference result from each individual model to get a better result.
AB - Medical image segmentation is the task of objective segmentation in medical field. 3D Tumor segmentation can help physicians efficiently diagnose cancer, track tumor change, and make treatment plans. With the development of machine learning (ML)/Deep Learning (DL) image segmentation methods, the performance of medical image segmentation has significantly improved especially in terms of accuracy and time efficiency. Performance of typical deep learning algorithms such as Fully Connection Networks, Unet, DeepLab varies with respect to different datasets, pre-processing and training parameter settings. In this paper, we propose a new architecture which utilizes the advantages of various models and aggregates their results. The original concept was inspired by Ensembles of Multiple Models and Architectures. In this paper, we train different sub-models separately. Then we train a gating network to credit the inference result from each individual model to get a better result.
KW - BraTS challenge
KW - Ensemble learning
KW - Machine learning
KW - Medical image segmentation
UR - http://www.scopus.com/inward/record.url?scp=85135162046&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85135162046&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-09002-8_13
DO - 10.1007/978-3-031-09002-8_13
M3 - Conference contribution
AN - SCOPUS:85135162046
SN - 9783031090011
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 138
EP - 148
BT - Brainlesion
A2 - Crimi, Alessandro
A2 - Bakas, Spyridon
PB - Springer Science and Business Media Deutschland GmbH
T2 - 7th International Brain Lesion Workshop, BrainLes 2021, held in conjunction with the Medical Image Computing and Computer Assisted Intervention, MICCAI 2021
Y2 - 27 September 2021 through 27 September 2021
ER -