eprintid: 15837
rev_number: 2
eprint_status: archive
userid: 1
dir: disk0/00/01/58/37
datestamp: 2023-11-10 03:30:28
lastmod: 2023-11-10 03:30:28
status_changed: 2023-11-10 02:00:32
type: article
metadata_visibility: show
creators_name: Khan, Z.
creators_name: Yahya, N.
creators_name: Alsaih, K.
creators_name: Meriaudeau, F.
title: Segmentation of Prostate in MRI Images Using Depth Separable Convolution Operations
ispublished: pub
keywords: Calcium compounds; Convolution; Decoding; Deep neural networks; Diseases; Electronic assessment; Human computer interaction; Image segmentation; Signal encoding; Urology, Computer-based assessments; Image preprocessing; Important features; Model performance; NET architecture; Prostate cancers; Prostate segmentation; Similarity coefficients, Magnetic resonance imaging
note: cited By 2; Conference of 12th International Conference on Intelligent Human Computer Interaction, IHCI 2020 ; Conference Date: 24 November 2020 Through 26 November 2020; Conference Code:255179
abstract: The segmentation of the prostate gland into two sub-regions, namely, the central gland (CG) and the peripheral zone (PZ) is crucial for the prostate cancer (PCa) diagnosis. The nature and occurrence of cancer occurred in the prostate is substantially different in both zones. Magnetic resonance imaging modality (MRI) is a clinically primary tool for computer-based assessment and remediation of various cancer types such as PCa. In this paper, we evaluated DeeplabV3+ model on T2W MRI scans using the I2CVB dataset, which is designed in an encoder-decoder style for the zonal segmentation of prostate regions. An important feature of DeeplabV3+ is the depth-wise separable convolutions, which allow more information to be extracted from images as it uses filters with different dilation rates. Prior to being fed to the deep neural network, image pre-processing techniques are applied, including image resizing, cropping, and denoising. The DeeplabV3+ model performance is evaluated using the Dice similarity coefficient (DSC) metric and compared with the vanilla U-Net architecture. Results show that the encoder-decoder network having depth-wise separable convolutions performed better prostate segmentation than the network with standard convolution operations with the DSC value of 70.1 in PZ and 81.5 in CG zone. © 2021, Springer Nature Switzerland AG.
date: 2021
publisher: Springer Science and Business Media Deutschland GmbH
official_url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85102291806&doi=10.1007%2f978-3-030-68449-5_14&partnerID=40&md5=f06c688949a24ac6b1ed937632ec5e48
id_number: 10.1007/978-3-030-68449-5₁₄
full_text_status: none
publication: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
volume: 12615 
pagerange: 132-141
refereed: TRUE
isbn: 9783030684488
issn: 03029743
citation:   Khan, Z. and Yahya, N. and Alsaih, K. and Meriaudeau, F.  (2021) Segmentation of Prostate in MRI Images Using Depth Separable Convolution Operations.  Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 12615 .  pp. 132-141.  ISSN 03029743