Alzyadat, Tariq and Praet, Stephan and Chetty, Girija and Goecke, Roland and Kumar, Dinesh and Welvaert, Marijke and Vlahovich, Nicole and Waddington, Gordon (2020) Automatic Segmentation of Achilles Tendon Tissues Using Deep Convolutional Neural Network. [Conference Proceedings]
Full text not available from this repository.Abstract
The automatic segmentation of Achilles tendon tissues is one of the preliminary steps towards creating a tool for diagnosing, prognosing, or monitoring changes in tendon organization over time. Manual delineation is the current approach of identifying Achilles region-of-interest (ROI), it is a tedious and time-consuming task. In this respect, the current work describes the first steps taken towards creating an automatic approach for Achilles tendon segmentation that utilize the capabilities of Deep Convolutional Neural Networks (CNNs). Firstly, the dataset has been pre-processed and manually segmented to be used as the ground-truth in the training and testing of the proposed automated model. Secondly, the model was trained and validated using three CNN architectures SegNet, ResNet-18 and ResNet-50. Finally, Tversky loss function, 3D augmentation and network ensembling approaches were used to improve the segmentation performance and to tackle challenges such as the limited size of the training dataset and data imbalance. The proposed fully automated segmentation method reached average Dice score of 0.904. In conclusion, this novel study demonstrates that a CNN approach is useful for performing accurate Achilles tendon segmentation in musculoskeletal imaging.
| Item Type: | Conference Proceedings |
|---|---|
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
| Divisions: | Faculty of Science, Technology and Environment (FSTE) > School of Computing, Information and Mathematical Sciences |
| Depositing User: | Dinesh Kumar |
| Date Deposited: | 27 Nov 2020 02:48 |
| Last Modified: | 05 Jul 2021 03:25 |
| URI: | http://repository.usp.ac.fj/id/eprint/12426 |
Actions (login required)
 |
View Item |