Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI
In this paper, we develop an optimised state-of-the-art 2D U-Net model by studying the effects of the individual deep learning model components in performing prostate segmentation. We found that for upsampling, the combination of interpolation and convolution is better than the use of transposed con...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-04-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/10/7/2601 |
| _version_ | 1797571050750345216 |
|---|---|
| author | Indriani P. Astono James S. Welsh Stephan Chalup Peter Greer |
| author_facet | Indriani P. Astono James S. Welsh Stephan Chalup Peter Greer |
| author_sort | Indriani P. Astono |
| collection | DOAJ |
| description | In this paper, we develop an optimised state-of-the-art 2D U-Net model by studying the effects of the individual deep learning model components in performing prostate segmentation. We found that for upsampling, the combination of interpolation and convolution is better than the use of transposed convolution. For combining feature maps in each convolution block, it is only beneficial if a skip connection with concatenation is used. With respect to pooling, average pooling is better than strided-convolution, max, RMS or L2 pooling. Introducing a batch normalisation layer before the activation layer gives further performance improvement. The optimisation is based on a private dataset as it has a fixed 2D resolution and voxel size for every image which mitigates the need of a resizing operation in the data preparation process. Non-enhancing data preprocessing was applied and five-fold cross-validation was used to evaluate the fully automatic segmentation approach. We show it outperforms the traditional methods that were previously applied on the private dataset, as well as outperforming other comparable state-of-the-art 2D models on the public dataset PROMISE12. |
| first_indexed | 2024-03-10T20:34:05Z |
| format | Article |
| id | doaj.art-a0560b7d686541fb82465d24bb747fed |
| institution | Directory Open Access Journal |
| issn | 2076-3417 |
| language | English |
| last_indexed | 2024-03-10T20:34:05Z |
| publishDate | 2020-04-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Applied Sciences |
| spelling | doaj.art-a0560b7d686541fb82465d24bb747fed2023-11-19T21:10:42ZengMDPI AGApplied Sciences2076-34172020-04-01107260110.3390/app10072601Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRIIndriani P. Astono0James S. Welsh1Stephan Chalup2Peter Greer3School of Electrical Engineering and Computing, The University of Newcastle, Callaghan, NSW 2308, AustraliaSchool of Electrical Engineering and Computing, The University of Newcastle, Callaghan, NSW 2308, AustraliaSchool of Electrical Engineering and Computing, The University of Newcastle, Callaghan, NSW 2308, AustraliaSchool of Mathematical and Physical Sciences, The University of Newcastle, Callaghan, NSW 2308, AustraliaIn this paper, we develop an optimised state-of-the-art 2D U-Net model by studying the effects of the individual deep learning model components in performing prostate segmentation. We found that for upsampling, the combination of interpolation and convolution is better than the use of transposed convolution. For combining feature maps in each convolution block, it is only beneficial if a skip connection with concatenation is used. With respect to pooling, average pooling is better than strided-convolution, max, RMS or L2 pooling. Introducing a batch normalisation layer before the activation layer gives further performance improvement. The optimisation is based on a private dataset as it has a fixed 2D resolution and voxel size for every image which mitigates the need of a resizing operation in the data preparation process. Non-enhancing data preprocessing was applied and five-fold cross-validation was used to evaluate the fully automatic segmentation approach. We show it outperforms the traditional methods that were previously applied on the private dataset, as well as outperforming other comparable state-of-the-art 2D models on the public dataset PROMISE12.https://www.mdpi.com/2076-3417/10/7/2601convolutional neural networksmedical image applicationprostate segmentationmagnetic resonance imagingMRI |
| spellingShingle | Indriani P. Astono James S. Welsh Stephan Chalup Peter Greer Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI Applied Sciences convolutional neural networks medical image application prostate segmentation magnetic resonance imaging MRI |
| title | Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI |
| title_full | Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI |
| title_fullStr | Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI |
| title_full_unstemmed | Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI |
| title_short | Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI |
| title_sort | optimisation of 2d u net model components for automatic prostate segmentation on mri |
| topic | convolutional neural networks medical image application prostate segmentation magnetic resonance imaging MRI |
| url | https://www.mdpi.com/2076-3417/10/7/2601 |
| work_keys_str_mv | AT indrianipastono optimisationof2dunetmodelcomponentsforautomaticprostatesegmentationonmri AT jamesswelsh optimisationof2dunetmodelcomponentsforautomaticprostatesegmentationonmri AT stephanchalup optimisationof2dunetmodelcomponentsforautomaticprostatesegmentationonmri AT petergreer optimisationof2dunetmodelcomponentsforautomaticprostatesegmentationonmri |