TY - GEN
T1 - Evaluation and comparison of global-feature-based and local-feature-based segmentation algorithms in intracranial visual pathway delineation
AU - Liu, Yan
AU - Gu, Xuejun
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/7
Y1 - 2020/7
N2 - Intracranial visual pathway is related to the effective transmission of visual signals to brain. It was not only the target organ of diseases but also the organs at risk in radiotherapy thus its delineation plays an important role in both diagnosis and treatment planning. Traditional manual segmentation method suffered from time- and labor- consuming as well as intra- and inter- variability. In order to overcome these problems, state-of-the-art segmentation models were designed and various features were extracted and utilized, but it's hard to tell their effectiveness on intracranial visual pathway delineation. It's because that these methods worked on different dataset and accompanied with different training tricks. This study aimed to research the contribution of global features and local features in delineating the intracranial visual pathway from MRI scans. The two typical segmentation models, 3D UNet and DeepMedic, were chosen since they focused on global features and local features respectively. We constructed the hybrid model through serially connecting the two mentioned models to validate the performance of combined global and local features. Validation results showed that the hybrid model outperformed the individual ones. It proved that multi scale feature fusion was important in improving the segmentation performance.
AB - Intracranial visual pathway is related to the effective transmission of visual signals to brain. It was not only the target organ of diseases but also the organs at risk in radiotherapy thus its delineation plays an important role in both diagnosis and treatment planning. Traditional manual segmentation method suffered from time- and labor- consuming as well as intra- and inter- variability. In order to overcome these problems, state-of-the-art segmentation models were designed and various features were extracted and utilized, but it's hard to tell their effectiveness on intracranial visual pathway delineation. It's because that these methods worked on different dataset and accompanied with different training tricks. This study aimed to research the contribution of global features and local features in delineating the intracranial visual pathway from MRI scans. The two typical segmentation models, 3D UNet and DeepMedic, were chosen since they focused on global features and local features respectively. We constructed the hybrid model through serially connecting the two mentioned models to validate the performance of combined global and local features. Validation results showed that the hybrid model outperformed the individual ones. It proved that multi scale feature fusion was important in improving the segmentation performance.
UR - http://www.scopus.com/inward/record.url?scp=85091026646&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85091026646&partnerID=8YFLogxK
U2 - 10.1109/EMBC44109.2020.9175937
DO - 10.1109/EMBC44109.2020.9175937
M3 - Conference contribution
C2 - 33018340
AN - SCOPUS:85091026646
T3 - Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS
SP - 1766
EP - 1769
BT - 42nd Annual International Conferences of the IEEE Engineering in Medicine and Biology Society
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 42nd Annual International Conferences of the IEEE Engineering in Medicine and Biology Society, EMBC 2020
Y2 - 20 July 2020 through 24 July 2020
ER -