Zonal Segmentation of Prostate T2W-MRI using Atrous Convolutional Neural Network

Khan, Z. and Yahya, N. and Alsaih, K. and Meriaudeau, F. (2019) Zonal Segmentation of Prostate T2W-MRI using Atrous Convolutional Neural Network. In: UNSPECIFIED.

Full text not available from this repository.
Official URL: https://www.scopus.com/inward/record.uri?eid=2-s2....

Abstract

The number of prostate cancer cases is steadily increasing especially with rising number of ageing population. It is reported that 5-year relative survival rate for man with stage 1 prostate cancer is almost 99 hence, early detection will significantly improve treatment planning and increase survival rate. Magnetic resonance imaging (MRI) technique is a common imaging modality for diagnosis of prostate cancer. MRI provide good visualization of soft tissue and enable better lesion detection and staging of prostate cancer. The main challenge of prostate whole gland segmentation is due to blurry boundary of central gland (CG) and peripheral zone (PZ) which lead to differential diagnosis. Since there is substantial difference in occurance and characteristic of cancer in both zones. So to enhance the diagnosis of prostate gland, we implemented DeeplabV3+ semantic segmentation approach to segment the prostate into zones. DeepLabV3+ achieved significant results in segmentation of prostate MRI by applying several parallel atrous convolution with different rates. The CNN-based semantic segmentation approach is trained and tested on NCI-ISBI 1.5T and 3T MRI dataset consist of 40 patients. Performance evaluation based on Dice similarity coefficient (DSC) of the Deeplab-based segmentation is compared with two other CNN-based semantic segmentation techniques: FCN and PSNet. Results shows that prostate segmentation using DeepLabV3+ can perform better than FCN and PSNet with average DSC of 70.3 in PZ and 88 in CG zone. This indicates the significant contribution made by the atrous convolution layer, in producing better prostate segmentation result. © 2019 IEEE.

Item Type: Conference or Workshop Item (UNSPECIFIED)
Impact Factor: cited By 5
Uncontrolled Keywords: Convolution; Deep learning; Diagnosis; Diseases; Image segmentation; Magnetic resonance imaging; Neural networks; Semantics, Convolutional neural network; DeepLabV3; Differential diagnosis; Gland segmentations; Prostate segmentation; PSNet; Semantic segmentation; Similarity coefficients, Urology
Depositing User: Ms Sharifah Fahimah Saiyed Yeop
Date Deposited: 19 Aug 2021 08:09
Last Modified: 19 Aug 2021 08:09
URI: http://scholars.utp.edu.my/id/eprint/23609

Actions (login required)

View Item
View Item